feat: add OI derived features (oi_change_ma5, oi_price_spread)

CLAUDE.md

@@ -47,7 +47,7 @@ bash scripts/deploy_model.sh
 **5-layer data flow on each 15m candle close:**
 1. `src/data_stream.py` — Combined WebSocket for XRP/BTC/ETH, deque buffers (200 candles each)
 2. `src/indicators.py` — RSI, MACD, BB, EMA, StochRSI, ATR; weighted signal aggregation → LONG/SHORT/HOLD
-3. `src/ml_filter.py` + `src/ml_features.py` — 24-feature extraction (ADX 포함), ONNX priority > LightGBM fallback, threshold ≥ 0.55
+3. `src/ml_filter.py` + `src/ml_features.py` — 26-feature extraction (ADX + OI 파생 피처 포함), ONNX priority > LightGBM fallback, threshold ≥ 0.55
 4. `src/exchange.py` + `src/risk_manager.py` — Dynamic margin, MARKET orders with SL/TP, daily loss limit (5%)
 5. `src/user_data_stream.py` + `src/notifier.py` — Real-time TP/SL detection via WebSocket, Discord webhooks
 
@@ -116,3 +116,4 @@ All design documents and implementation plans are stored in `docs/plans/` with t
 | 2026-03-03 | `position-monitor-logging` | Completed |
 | 2026-03-03 | `adx-ml-feature-migration` (design + plan) | Completed |
 | 2026-03-03 | `optuna-precision-objective-plan` | Pending |
+| 2026-03-04 | `oi-derived-features` (design + plan) | In Progress |

docs/plans/2026-03-04-oi-derived-features-design.md

@@ -0,0 +1,70 @@
+# OI 파생 피처 설계
+
+## 목표
+
+기존 `oi_change` 피처에 더해, OI 데이터에서 파생 피처 2개를 만들어 LightGBM 학습 데이터에 추가하고, 피처 추가 전후 검증셋 성능을 자동 비교한다.
+
+## 제약사항
+
+- Binance OI 히스토리 API는 최근 30일분만 제공
+- 학습 데이터에서 OI 유효 구간 ≈ 2,880개 15분 캔들
+- A/B 비교 결과는 방향성 참고용 (통계적 유의성 제한)
+
+## 파생 피처
+
+### 1. `oi_change_ma5`
+
+- **계산**: OI 변화율의 5캔들(75분) 이동평균
+- **의미**: OI 단기 추세. 급감/급증 노이즈 제거된 방향성
+- **정규화**: rolling z-score (288캔들 윈도우, 기존 패턴 동일)
+- **기존 `oi_change`와의 관계**: smoothed 버전. 상관관계 높을 수 있으나 LightGBM이 자연 선택. importance 낮으면 이후 제거
+
+### 2. `oi_price_spread`
+
+- **계산**: `rolling_zscore(oi_change) - rolling_zscore(price_ret_1)`
+- **의미**: OI와 가격 움직임 간 괴리도 (연속값)
+  - 양수: OI가 가격 대비 강세 (자금 유입)
+  - 음수: OI가 가격 대비 약세 (자금 유출)
+- **정규화**: 양쪽 입력이 이미 z-score이므로 추가 정규화 불필요
+- **바이너리 대신 연속값 채택 이유**: sign() 기반 바이너리는 미미한 차이도 1/0으로 분류 → 노이즈 과잉. 연속값은 LightGBM이 분할점을 학습
+
+## 수정 대상 파일
+
+### dataset_builder.py
+
+- OI 파생 피처 2개 계산 로직 추가
+- 기존 `oi_change` z-score 결과를 재사용하여 `oi_change_ma5` 계산
+- `oi_price_spread` = `oi_change` z-score - `ret_1` z-score
+
+### ml_features.py
+
+- `FEATURE_COLS`에 `oi_change_ma5`, `oi_price_spread` 추가 (24→26)
+- `build_features()`에 실시간 계산 로직 추가
+  - `oi_change_ma5`: bot에서 전달받은 최근 5봉 OI MA
+  - `oi_price_spread`: 실시간 z-scored OI - z-scored price change
+
+### train_model.py
+
+- `--compare` 플래그 추가
+- Baseline (기존 24피처) vs New (26피처) 자동 비교 출력:
+  - Precision, Recall, F1, AUC-ROC
+  - Feature importance top 10
+  - Best threshold
+  - 검증셋 크기 (n=XX) 및 "방향성 참고용" 경고
+
+### bot.py
+
+- OI 변화율 히스토리 deque(maxlen=5) 관리
+- `_init_oi_history()`: 봇 시작 시 Binance OI hist API에서 최근 5봉 fetch → cold start 해결
+- `_fetch_market_microstructure()` 확장: MA5 계산, price_spread 계산 후 build_features()에 전달
+
+### exchange.py
+
+- `get_oi_history(limit=5)`: 봇 초기화용 최근 OI 히스토리 fetch 메서드 추가
+
+### scripts/collect_oi.py (신규)
+
+- OI 장기 수집 스크립트
+- 15분마다 cron 실행
+- Binance `/fapi/v1/openInterest` 호출 → `data/oi_history.parquet`에 append
+- 기존 fetch_history.py의 30일 데이터 보완용

docs/plans/2026-03-04-oi-derived-features-plan.md

@@ -0,0 +1,764 @@
+# OI 파생 피처 구현 계획
+
+> **For Claude:** REQUIRED SUB-SKILL: Use superpowers:executing-plans to implement this plan task-by-task.
+
+**Goal:** OI 파생 피처 2개(`oi_change_ma5`, `oi_price_spread`)를 추가하고, 기존 대비 성능을 자동 비교하며, OI 장기 수집 스크립트를 만든다.
+
+**Architecture:** dataset_builder.py에 파생 피처 계산 추가 → ml_features.py에 FEATURE_COLS/build_features 확장 → train_model.py에 --compare 플래그로 A/B 비교 → bot.py에 OI deque 히스토리 관리 및 cold start → scripts/collect_oi.py 신규
+
+**Tech Stack:** Python, LightGBM, pandas, numpy, Binance REST API
+
+---
+
+### Task 1: dataset_builder.py — OI 파생 피처 계산
+
+**Files:**
+- Modify: `src/dataset_builder.py:277-291` (OI/FR 피처 계산 블록)
+- Test: `tests/test_dataset_builder.py`
+
+**Step 1: Write failing tests**
+
+`tests/test_dataset_builder.py` 끝에 추가:
+
+```python
+def test_oi_derived_features_present():
+    """OI 파생 피처 2개가 결과에 포함되어야 한다."""
+    import numpy as np
+    import pandas as pd
+    from src.dataset_builder import _calc_features_vectorized, _calc_signals, _calc_indicators
+
+    n = 300
+    np.random.seed(42)
+    df = pd.DataFrame({
+        "open":      np.random.uniform(1, 2, n),
+        "high":      np.random.uniform(2, 3, n),
+        "low":       np.random.uniform(0.5, 1, n),
+        "close":     np.random.uniform(1, 2, n),
+        "volume":    np.random.uniform(1000, 5000, n),
+        "oi_change": np.concatenate([np.zeros(100), np.random.uniform(-0.05, 0.05, 200)]),
+    })
+    d = _calc_indicators(df)
+    sig = _calc_signals(d)
+    feat = _calc_features_vectorized(d, sig)
+
+    assert "oi_change_ma5" in feat.columns, "oi_change_ma5 컬럼이 없음"
+    assert "oi_price_spread" in feat.columns, "oi_price_spread 컬럼이 없음"
+
+
+def test_oi_derived_features_nan_when_no_oi():
+    """oi_change 컬럼이 없으면 파생 피처도 nan이어야 한다."""
+    import numpy as np
+    import pandas as pd
+    from src.dataset_builder import _calc_features_vectorized, _calc_signals, _calc_indicators
+
+    n = 200
+    np.random.seed(0)
+    df = pd.DataFrame({
+        "open":   np.random.uniform(1, 2, n),
+        "high":   np.random.uniform(2, 3, n),
+        "low":    np.random.uniform(0.5, 1, n),
+        "close":  np.random.uniform(1, 2, n),
+        "volume": np.random.uniform(1000, 5000, n),
+    })
+    d = _calc_indicators(df)
+    sig = _calc_signals(d)
+    feat = _calc_features_vectorized(d, sig)
+
+    assert feat["oi_change_ma5"].isna().all(), "oi_change 컬럼 없을 때 oi_change_ma5는 전부 nan이어야 함"
+    assert feat["oi_price_spread"].isna().all(), "oi_change 컬럼 없을 때 oi_price_spread는 전부 nan이어야 함"
+
+
+def test_oi_price_spread_is_continuous():
+    """oi_price_spread는 바이너리가 아닌 연속값이어야 한다."""
+    import numpy as np
+    import pandas as pd
+    from src.dataset_builder import _calc_features_vectorized, _calc_signals, _calc_indicators
+
+    n = 300
+    np.random.seed(42)
+    df = pd.DataFrame({
+        "open":      np.random.uniform(1, 2, n),
+        "high":      np.random.uniform(2, 3, n),
+        "low":       np.random.uniform(0.5, 1, n),
+        "close":     np.random.uniform(1, 2, n),
+        "volume":    np.random.uniform(1000, 5000, n),
+        "oi_change": np.random.uniform(-0.05, 0.05, n),
+    })
+    d = _calc_indicators(df)
+    sig = _calc_signals(d)
+    feat = _calc_features_vectorized(d, sig)
+
+    valid = feat["oi_price_spread"].dropna()
+    assert len(valid.unique()) > 2, "oi_price_spread는 연속값이어야 함 (2개 초과 유니크 값)"
+```
+
+**Step 2: Run tests to verify they fail**
+
+Run: `bash scripts/run_tests.sh -k "oi_derived"`
+Expected: FAIL — `oi_change_ma5`, `oi_price_spread` 컬럼 없음
+
+**Step 3: Implement in dataset_builder.py**
+
+`src/dataset_builder.py:277-291` (기존 OI/FR 블록) 뒤에 파생 피처 추가:
+
+```python
+    # OI 변화율 / 펀딩비 피처
+    # 컬럼 없으면 전체 nan, 있으면 0.0 구간(데이터 미제공 구간)을 nan으로 마스킹
+    if "oi_change" in d.columns:
+        oi_raw = np.where(d["oi_change"].values == 0.0, np.nan, d["oi_change"].values)
+    else:
+        oi_raw = np.full(len(d), np.nan)
+
+    if "funding_rate" in d.columns:
+        fr_raw = np.where(d["funding_rate"].values == 0.0, np.nan, d["funding_rate"].values)
+    else:
+        fr_raw = np.full(len(d), np.nan)
+
+    oi_z = _rolling_zscore(oi_raw.astype(np.float64), window=96)
+    result["oi_change"]    = oi_z
+    result["funding_rate"] = _rolling_zscore(fr_raw.astype(np.float64), window=96)
+
+    # --- OI 파생 피처 ---
+    # 1. oi_change_ma5: OI 변화율의 5캔들 이동평균 (단기 추세)
+    oi_series = pd.Series(oi_raw.astype(np.float64))
+    oi_ma5_raw = oi_series.rolling(window=5, min_periods=1).mean().values
+    result["oi_change_ma5"] = _rolling_zscore(oi_ma5_raw, window=96)
+
+    # 2. oi_price_spread: z-scored OI 변화율 - z-scored 가격 수익률 (연속값)
+    #    양수: OI가 가격 대비 강세 (자금 유입)
+    #    음수: OI가 가격 대비 약세 (자금 유출)
+    result["oi_price_spread"] = oi_z - ret_1_z
+```
+
+주의: 기존 `oi_change`와 `funding_rate`의 window도 288→96으로 변경. `oi_z` 변수를 재사용하여 `oi_price_spread` 계산. `ret_1_z`는 이미 위에서 계산됨 (line 181).
+
+**Step 4: Update OPTIONAL_COLS in generate_dataset_vectorized**
+
+`src/dataset_builder.py:387` 수정:
+
+```python
+    OPTIONAL_COLS = {"oi_change", "funding_rate", "oi_change_ma5", "oi_price_spread"}
+```
+
+**Step 5: Run tests to verify they pass**
+
+Run: `bash scripts/run_tests.sh -k "oi_derived"`
+Expected: 3 tests PASS
+
+**Step 6: Run full test suite**
+
+Run: `bash scripts/run_tests.sh`
+Expected: All existing tests PASS (기존 oi_change/funding_rate 테스트 포함)
+
+**Step 7: Commit**
+
+```bash
+git add src/dataset_builder.py tests/test_dataset_builder.py
+git commit -m "feat: add oi_change_ma5 and oi_price_spread derived features to dataset builder"
+```
+
+---
+
+### Task 2: ml_features.py — FEATURE_COLS 및 build_features() 확장
+
+**Files:**
+- Modify: `src/ml_features.py:4-15` (FEATURE_COLS), `src/ml_features.py:33-139` (build_features)
+- Test: `tests/test_ml_features.py`
+
+**Step 1: Write failing tests**
+
+`tests/test_ml_features.py` 끝에 추가:
+
+```python
+def test_feature_cols_has_26_items():
+    from src.ml_features import FEATURE_COLS
+    assert len(FEATURE_COLS) == 26
+
+
+def test_build_features_with_oi_derived_params():
+    """oi_change_ma5, oi_price_spread 파라미터가 피처에 반영된다."""
+    xrp_df = _make_df(10, base_price=1.0)
+    btc_df = _make_df(10, base_price=50000.0)
+    eth_df = _make_df(10, base_price=3000.0)
+    features = build_features(
+        xrp_df, "LONG",
+        btc_df=btc_df, eth_df=eth_df,
+        oi_change=0.05, funding_rate=0.0002,
+        oi_change_ma5=0.03, oi_price_spread=0.12,
+    )
+    assert features["oi_change_ma5"] == pytest.approx(0.03)
+    assert features["oi_price_spread"] == pytest.approx(0.12)
+
+
+def test_build_features_oi_derived_defaults_to_zero():
+    """oi_change_ma5, oi_price_spread 미제공 시 0.0으로 채워진다."""
+    xrp_df = _make_df(10, base_price=1.0)
+    features = build_features(xrp_df, "LONG")
+    assert features["oi_change_ma5"] == pytest.approx(0.0)
+    assert features["oi_price_spread"] == pytest.approx(0.0)
+```
+
+기존 테스트 수정:
+- `test_feature_cols_has_24_items` → 삭제 또는 숫자를 26으로 변경
+- `test_build_features_with_btc_eth_has_24_features` → `assert len(features) == 26`
+- `test_build_features_without_btc_eth_has_16_features` → `assert len(features) == 18`
+
+**Step 2: Run tests to verify they fail**
+
+Run: `bash scripts/run_tests.sh -k "test_feature_cols_has_26 or test_build_features_oi_derived"`
+Expected: FAIL
+
+**Step 3: Implement**
+
+`src/ml_features.py` FEATURE_COLS 수정 (line 4-15):
+
+```python
+FEATURE_COLS = [
+    "rsi", "macd_hist", "bb_pct", "ema_align",
+    "stoch_k", "stoch_d", "atr_pct", "vol_ratio",
+    "ret_1", "ret_3", "ret_5", "signal_strength", "side",
+    "btc_ret_1", "btc_ret_3", "btc_ret_5",
+    "eth_ret_1", "eth_ret_3", "eth_ret_5",
+    "xrp_btc_rs", "xrp_eth_rs",
+    # 시장 미시구조: OI 변화율(z-score), 펀딩비(z-score)
+    "oi_change", "funding_rate",
+    # OI 파생 피처
+    "oi_change_ma5", "oi_price_spread",
+    "adx",
+]
+```
+
+`build_features()` 시그니처 수정 (line 33-40):
+
+```python
+def build_features(
+    df: pd.DataFrame,
+    signal: str,
+    btc_df: pd.DataFrame | None = None,
+    eth_df: pd.DataFrame | None = None,
+    oi_change: float | None = None,
+    funding_rate: float | None = None,
+    oi_change_ma5: float | None = None,
+    oi_price_spread: float | None = None,
+) -> pd.Series:
+```
+
+`build_features()` 끝부분 (line 134-138) 수정:
+
+```python
+    base["oi_change"]       = float(oi_change)       if oi_change       is not None else 0.0
+    base["funding_rate"]    = float(funding_rate)    if funding_rate    is not None else 0.0
+    base["oi_change_ma5"]   = float(oi_change_ma5)   if oi_change_ma5   is not None else 0.0
+    base["oi_price_spread"] = float(oi_price_spread) if oi_price_spread is not None else 0.0
+    base["adx"] = float(last.get("adx", 0))
+```
+
+**Step 4: Run tests**
+
+Run: `bash scripts/run_tests.sh -k "test_ml_features"`
+Expected: All PASS
+
+**Step 5: Run full test suite**
+
+Run: `bash scripts/run_tests.sh`
+Expected: All PASS (test_dataset_builder의 FEATURE_COLS 참조도 26개로 통과)
+
+**Step 6: Commit**
+
+```bash
+git add src/ml_features.py tests/test_ml_features.py
+git commit -m "feat: add oi_change_ma5 and oi_price_spread to FEATURE_COLS and build_features"
+```
+
+---
+
+### Task 3: train_model.py — --compare A/B 비교 모드
+
+**Files:**
+- Modify: `scripts/train_model.py:425-452` (main, argparse)
+- Test: 수동 실행 확인 (학습 스크립트는 통합 테스트)
+
+**Step 1: Implement compare function**
+
+`scripts/train_model.py`에 `compare()` 함수 추가 (train() 함수 뒤):
+
+```python
+def compare(data_path: str, time_weight_decay: float = 2.0, tuned_params_path: str | None = None):
+    """기존 피처 vs OI 파생 피처 추가 버전 A/B 비교."""
+    print("=" * 70)
+    print("  OI 파생 피처 A/B 비교 (30일 데이터 기반, 방향성 참고용)")
+    print("=" * 70)
+
+    df_raw = pd.read_parquet(data_path)
+    base_cols = ["open", "high", "low", "close", "volume"]
+    btc_df = eth_df = None
+    if "close_btc" in df_raw.columns:
+        btc_df = df_raw[[c + "_btc" for c in base_cols]].copy()
+        btc_df.columns = base_cols
+    if "close_eth" in df_raw.columns:
+        eth_df = df_raw[[c + "_eth" for c in base_cols]].copy()
+        eth_df.columns = base_cols
+    df = df_raw[base_cols].copy()
+    if "oi_change" in df_raw.columns:
+        df["oi_change"] = df_raw["oi_change"]
+    if "funding_rate" in df_raw.columns:
+        df["funding_rate"] = df_raw["funding_rate"]
+
+    dataset = generate_dataset_vectorized(
+        df, btc_df=btc_df, eth_df=eth_df,
+        time_weight_decay=time_weight_decay,
+        negative_ratio=5,
+    )
+
+    if dataset.empty:
+        raise ValueError("데이터셋 생성 실패")
+
+    lgbm_params, weight_scale = _load_lgbm_params(tuned_params_path)
+
+    # Baseline: OI 파생 피처 제외
+    BASELINE_EXCLUDE = {"oi_change_ma5", "oi_price_spread"}
+    baseline_cols = [c for c in FEATURE_COLS if c in dataset.columns and c not in BASELINE_EXCLUDE]
+    new_cols = [c for c in FEATURE_COLS if c in dataset.columns]
+
+    results = {}
+    for label, cols in [("Baseline (24)", baseline_cols), ("New (26)", new_cols)]:
+        X = dataset[cols]
+        y = dataset["label"]
+        w = dataset["sample_weight"].values
+        source = dataset["source"].values if "source" in dataset.columns else np.full(len(X), "signal")
+
+        split = int(len(X) * 0.8)
+        X_tr, X_val = X.iloc[:split], X.iloc[split:]
+        y_tr, y_val = y.iloc[:split], y.iloc[split:]
+        w_tr = (w[:split] * weight_scale).astype(np.float32)
+        source_tr = source[:split]
+
+        balanced_idx = stratified_undersample(y_tr.values, source_tr, seed=42)
+        X_tr_b = X_tr.iloc[balanced_idx]
+        y_tr_b = y_tr.iloc[balanced_idx]
+        w_tr_b = w_tr[balanced_idx]
+
+        import warnings
+        model = lgb.LGBMClassifier(**lgbm_params, random_state=42, verbose=-1)
+        with warnings.catch_warnings():
+            warnings.simplefilter("ignore")
+            model.fit(X_tr_b, y_tr_b, sample_weight=w_tr_b)
+
+        proba = model.predict_proba(X_val)[:, 1]
+        auc = roc_auc_score(y_val, proba) if len(np.unique(y_val)) > 1 else 0.5
+
+        precs, recs, thrs = precision_recall_curve(y_val, proba)
+        precs, recs = precs[:-1], recs[:-1]
+        valid_idx = np.where(recs >= 0.15)[0]
+        if len(valid_idx) > 0:
+            best_i = valid_idx[np.argmax(precs[valid_idx])]
+            thr, prec, rec = float(thrs[best_i]), float(precs[best_i]), float(recs[best_i])
+        else:
+            thr, prec, rec = 0.50, 0.0, 0.0
+
+        # Feature importance
+        imp = dict(zip(cols, model.feature_importances_))
+        top10 = sorted(imp.items(), key=lambda x: x[1], reverse=True)[:10]
+
+        results[label] = {
+            "auc": auc, "precision": prec, "recall": rec,
+            "threshold": thr, "n_val": len(y_val),
+            "n_val_pos": int(y_val.sum()), "top10": top10,
+        }
+
+    # 비교 테이블 출력
+    print(f"\n{'지표':<20} {'Baseline (24)':>15} {'New (26)':>15} {'Delta':>10}")
+    print("-" * 62)
+    for metric in ["auc", "precision", "recall", "threshold"]:
+        b = results["Baseline (24)"][metric]
+        n = results["New (26)"][metric]
+        d = n - b
+        sign = "+" if d > 0 else ""
+        print(f"{metric:<20} {b:>15.4f} {n:>15.4f} {sign}{d:>9.4f}")
+
+    n_val = results["Baseline (24)"]["n_val"]
+    n_pos = results["Baseline (24)"]["n_val_pos"]
+    print(f"\n검증셋: n={n_val} (양성={n_pos}, 음성={n_val - n_pos})")
+    print("⚠ 30일 데이터 기반 — 방향성 참고용\n")
+
+    print("Feature Importance Top 10 (New):")
+    for feat_name, imp_val in results["New (26)"]["top10"]:
+        marker = " ← NEW" if feat_name in BASELINE_EXCLUDE else ""
+        print(f"  {feat_name:<25} {imp_val:>6}{marker}")
+```
+
+**Step 2: Add --compare flag to argparse**
+
+`scripts/train_model.py` main() 함수의 argparse에 추가:
+
+```python
+    parser.add_argument("--compare", action="store_true",
+                        help="OI 파생 피처 추가 전후 A/B 성능 비교")
+```
+
+main() 분기에 추가:
+
+```python
+    if args.compare:
+        compare(args.data, time_weight_decay=args.decay, tuned_params_path=args.tuned_params)
+    elif args.wf:
+        ...
+```
+
+**Step 3: Commit**
+
+```bash
+git add scripts/train_model.py
+git commit -m "feat: add --compare flag for OI derived features A/B comparison"
+```
+
+---
+
+### Task 4: bot.py — OI deque 히스토리 및 실시간 파생 피처 공급
+
+**Files:**
+- Modify: `src/bot.py:15-31` (init), `src/bot.py:60-83` (fetch/calc), `src/bot.py:110-114,287-291` (build_features 호출)
+- Modify: `src/exchange.py` (get_oi_history 추가)
+- Test: `tests/test_bot.py`
+
+**Step 1: Write failing tests**
+
+`tests/test_bot.py` 끝에 추가:
+
+```python
+def test_bot_has_oi_history_deque(config):
+    """봇이 OI 히스토리 deque를 가져야 한다."""
+    with patch("src.bot.BinanceFuturesClient"):
+        bot = TradingBot(config)
+    from collections import deque
+    assert isinstance(bot._oi_history, deque)
+    assert bot._oi_history.maxlen == 5
+
+
+@pytest.mark.asyncio
+async def test_init_oi_history_fills_deque(config):
+    """_init_oi_history가 deque를 채워야 한다."""
+    with patch("src.bot.BinanceFuturesClient"):
+        bot = TradingBot(config)
+    bot.exchange.get_oi_history = AsyncMock(return_value=[0.01, -0.02, 0.03, -0.01, 0.02])
+    await bot._init_oi_history()
+    assert len(bot._oi_history) == 5
+
+
+@pytest.mark.asyncio
+async def test_fetch_microstructure_returns_derived_features(config):
+    """_fetch_market_microstructure가 oi_change_ma5와 oi_price_spread를 반환해야 한다."""
+    with patch("src.bot.BinanceFuturesClient"):
+        bot = TradingBot(config)
+    bot.exchange.get_open_interest = AsyncMock(return_value=5000000.0)
+    bot.exchange.get_funding_rate = AsyncMock(return_value=0.0001)
+    bot._prev_oi = 4900000.0
+    bot._oi_history.extend([0.01, -0.02, 0.03, -0.01])
+    bot._latest_ret_1 = 0.01
+
+    result = await bot._fetch_market_microstructure()
+    assert len(result) == 4  # oi_change, funding_rate, oi_change_ma5, oi_price_spread
+```
+
+**Step 2: Run tests to verify they fail**
+
+Run: `bash scripts/run_tests.sh -k "oi_history or fetch_microstructure_returns_derived"`
+Expected: FAIL
+
+**Step 3: Implement exchange.get_oi_history()**
+
+`src/exchange.py`에 추가:
+
+```python
+    async def get_oi_history(self, limit: int = 5) -> list[float]:
+        """최근 OI 변화율 히스토리를 조회한다 (봇 초기화용). 실패 시 빈 리스트."""
+        loop = asyncio.get_event_loop()
+        try:
+            result = await loop.run_in_executor(
+                None,
+                lambda: self.client.futures_open_interest_hist(
+                    symbol=self.config.symbol, period="15m", limit=limit + 1,
+                ),
+            )
+            if len(result) < 2:
+                return []
+            oi_values = [float(r["sumOpenInterest"]) for r in result]
+            changes = []
+            for i in range(1, len(oi_values)):
+                if oi_values[i - 1] > 0:
+                    changes.append((oi_values[i] - oi_values[i - 1]) / oi_values[i - 1])
+                else:
+                    changes.append(0.0)
+            return changes
+        except Exception as e:
+            logger.warning(f"OI 히스토리 조회 실패 (무시): {e}")
+            return []
+```
+
+**Step 4: Implement bot.py changes**
+
+`src/bot.py` `__init__` 수정:
+
+```python
+from collections import deque
+
+# __init__에 추가:
+        self._oi_history: deque = deque(maxlen=5)
+        self._latest_ret_1: float = 0.0  # 최신 가격 수익률 (oi_price_spread용)
+```
+
+`_init_oi_history()` 추가:
+
+```python
+    async def _init_oi_history(self) -> None:
+        """봇 시작 시 최근 OI 변화율 히스토리를 조회하여 deque를 채운다."""
+        try:
+            changes = await self.exchange.get_oi_history(limit=5)
+            for c in changes:
+                self._oi_history.append(c)
+            if changes:
+                self._prev_oi = None  # 다음 실시간 OI로 갱신
+            logger.info(f"OI 히스토리 초기화: {len(self._oi_history)}개")
+        except Exception as e:
+            logger.warning(f"OI 히스토리 초기화 실패 (무시): {e}")
+```
+
+`_fetch_market_microstructure()` 수정 — 4-tuple 반환:
+
+```python
+    async def _fetch_market_microstructure(self) -> tuple[float, float, float, float]:
+        """OI 변화율, 펀딩비, OI MA5, OI-가격 스프레드를 실시간으로 조회한다."""
+        oi_val, fr_val = await asyncio.gather(
+            self.exchange.get_open_interest(),
+            self.exchange.get_funding_rate(),
+            return_exceptions=True,
+        )
+        if isinstance(oi_val, (int, float)) and oi_val > 0:
+            oi_change = self._calc_oi_change(float(oi_val))
+        else:
+            oi_change = 0.0
+        fr_float = float(fr_val) if isinstance(fr_val, (int, float)) else 0.0
+
+        # OI 히스토리 업데이트 및 MA5 계산
+        self._oi_history.append(oi_change)
+        oi_ma5 = sum(self._oi_history) / len(self._oi_history) if self._oi_history else 0.0
+
+        # OI-가격 스프레드 (단순 차이, 실시간에서는 z-score 없이 raw)
+        oi_price_spread = oi_change - self._latest_ret_1
+
+        logger.debug(
+            f"OI={oi_val}, OI변화율={oi_change:.6f}, 펀딩비={fr_float:.6f}, "
+            f"OI_MA5={oi_ma5:.6f}, OI_Price_Spread={oi_price_spread:.6f}"
+        )
+        return oi_change, fr_float, oi_ma5, oi_price_spread
+```
+
+`process_candle()` 수정:
+
+```python
+        # 캔들 마감 시 가격 수익률 계산 (oi_price_spread용)
+        if len(df) >= 2:
+            prev_close = df["close"].iloc[-2]
+            curr_close = df["close"].iloc[-1]
+            self._latest_ret_1 = (curr_close - prev_close) / prev_close if prev_close != 0 else 0.0
+
+        oi_change, funding_rate, oi_ma5, oi_price_spread = await self._fetch_market_microstructure()
+```
+
+모든 `build_features()` 호출에 새 파라미터 추가:
+
+```python
+            features = build_features(
+                df_with_indicators, signal,
+                btc_df=btc_df, eth_df=eth_df,
+                oi_change=oi_change, funding_rate=funding_rate,
+                oi_change_ma5=oi_ma5, oi_price_spread=oi_price_spread,
+            )
+```
+
+`_close_and_reenter()` 시그니처도 확장:
+
+```python
+    async def _close_and_reenter(
+        self,
+        position: dict,
+        signal: str,
+        df,
+        btc_df=None,
+        eth_df=None,
+        oi_change: float = 0.0,
+        funding_rate: float = 0.0,
+        oi_change_ma5: float = 0.0,
+        oi_price_spread: float = 0.0,
+    ) -> None:
+```
+
+`run()` 수정 — `_init_oi_history()` 호출 추가:
+
+```python
+    async def run(self):
+        logger.info(f"봇 시작: {self.config.symbol}, 레버리지 {self.config.leverage}x")
+        await self._recover_position()
+        await self._init_oi_history()
+        ...
+```
+
+**Step 5: Run tests**
+
+Run: `bash scripts/run_tests.sh -k "test_bot"`
+Expected: All PASS
+
+**Step 6: Run full test suite**
+
+Run: `bash scripts/run_tests.sh`
+Expected: All PASS
+
+**Step 7: Commit**
+
+```bash
+git add src/bot.py src/exchange.py tests/test_bot.py
+git commit -m "feat: add OI history deque, cold start init, and derived features to bot runtime"
+```
+
+---
+
+### Task 5: scripts/collect_oi.py — OI 장기 수집 스크립트
+
+**Files:**
+- Create: `scripts/collect_oi.py`
+
+**Step 1: Implement**
+
+```python
+"""
+OI 장기 수집 스크립트.
+15분마다 cron 실행하여 Binance OI를 data/oi_history.parquet에 누적한다.
+
+사용법:
+  python scripts/collect_oi.py
+  python scripts/collect_oi.py --symbol XRPUSDT
+
+crontab 예시:
+  */15 * * * * cd /path/to/cointrader && .venv/bin/python scripts/collect_oi.py >> logs/collect_oi.log 2>&1
+"""
+import sys
+from pathlib import Path
+sys.path.insert(0, str(Path(__file__).parent.parent))
+
+import argparse
+from datetime import datetime, timezone
+
+import pandas as pd
+from binance.client import Client
+from dotenv import load_dotenv
+import os
+
+load_dotenv()
+
+OI_PATH = Path("data/oi_history.parquet")
+
+
+def collect(symbol: str = "XRPUSDT"):
+    client = Client(
+        api_key=os.getenv("BINANCE_API_KEY", ""),
+        api_secret=os.getenv("BINANCE_API_SECRET", ""),
+    )
+
+    result = client.futures_open_interest(symbol=symbol)
+    oi_value = float(result["openInterest"])
+    ts = datetime.now(timezone.utc)
+
+    new_row = pd.DataFrame([{
+        "timestamp": ts,
+        "symbol": symbol,
+        "open_interest": oi_value,
+    }])
+
+    if OI_PATH.exists():
+        existing = pd.read_parquet(OI_PATH)
+        combined = pd.concat([existing, new_row], ignore_index=True)
+    else:
+        OI_PATH.parent.mkdir(parents=True, exist_ok=True)
+        combined = new_row
+
+    combined.to_parquet(OI_PATH, index=False)
+    print(f"[{ts.isoformat()}] OI={oi_value:.2f} → {OI_PATH}")
+
+
+def main():
+    parser = argparse.ArgumentParser(description="OI 장기 수집")
+    parser.add_argument("--symbol", default="XRPUSDT")
+    args = parser.parse_args()
+    collect(symbol=args.symbol)
+
+
+if __name__ == "__main__":
+    main()
+```
+
+**Step 2: Commit**
+
+```bash
+git add scripts/collect_oi.py
+git commit -m "feat: add OI long-term collection script for cron-based data accumulation"
+```
+
+---
+
+### Task 6: 기존 테스트 수정 및 전체 검증
+
+**Files:**
+- Modify: `tests/test_ml_features.py` (피처 수 변경)
+- Modify: `tests/test_bot.py` (기존 OI 테스트가 4-tuple 반환에 호환되도록)
+
+**Step 1: Fix test_ml_features.py assertions**
+
+- `test_feature_cols_has_24_items` → 26으로 변경
+- `test_build_features_with_btc_eth_has_24_features` → 26
+- `test_build_features_without_btc_eth_has_16_features` → 18
+
+**Step 2: Fix test_bot.py**
+
+기존 `test_process_candle_fetches_oi_and_funding` 등에서 `_fetch_market_microstructure` 반환값이 4-tuple이 되므로 mock 반환값 수정:
+
+```python
+bot._fetch_market_microstructure = AsyncMock(return_value=(0.02, 0.0001, 0.015, 0.01))
+```
+
+또는 `_fetch_market_microstructure`를 mock하지 않는 테스트는 exchange mock이 정상이면 자동 통과.
+
+**Step 3: Run full test suite**
+
+Run: `bash scripts/run_tests.sh`
+Expected: All PASS
+
+**Step 4: Commit**
+
+```bash
+git add tests/test_ml_features.py tests/test_bot.py
+git commit -m "test: update test assertions for 26-feature model and 4-tuple microstructure"
+```
+
+---
+
+### Task 7: CLAUDE.md 업데이트
+
+**Files:**
+- Modify: `CLAUDE.md`
+
+**Step 1: Update plan table**
+
+CLAUDE.md의 plan history 테이블에 추가:
+
+```
+| 2026-03-04 | `oi-derived-features` (design + plan) | In Progress |
+```
+
+ml_features.py 설명도 24→26개로 갱신.
+
+**Step 2: Commit**
+
+```bash
+git add CLAUDE.md
+git commit -m "docs: update CLAUDE.md with OI derived features plan status"
+```

scripts/collect_oi.py

@@ -0,0 +1,64 @@
+"""
+OI 장기 수집 스크립트.
+15분마다 cron 실행하여 Binance OI를 data/oi_history.parquet에 누적한다.
+
+사용법:
+  python scripts/collect_oi.py
+  python scripts/collect_oi.py --symbol XRPUSDT
+
+crontab 예시:
+  */15 * * * * cd /path/to/cointrader && .venv/bin/python scripts/collect_oi.py >> logs/collect_oi.log 2>&1
+"""
+import sys
+from pathlib import Path
+sys.path.insert(0, str(Path(__file__).parent.parent))
+
+import argparse
+from datetime import datetime, timezone
+
+import pandas as pd
+from binance.client import Client
+from dotenv import load_dotenv
+import os
+
+load_dotenv()
+
+OI_PATH = Path("data/oi_history.parquet")
+
+
+def collect(symbol: str = "XRPUSDT"):
+    client = Client(
+        api_key=os.getenv("BINANCE_API_KEY", ""),
+        api_secret=os.getenv("BINANCE_API_SECRET", ""),
+    )
+
+    result = client.futures_open_interest(symbol=symbol)
+    oi_value = float(result["openInterest"])
+    ts = datetime.now(timezone.utc)
+
+    new_row = pd.DataFrame([{
+        "timestamp": ts,
+        "symbol": symbol,
+        "open_interest": oi_value,
+    }])
+
+    if OI_PATH.exists():
+        existing = pd.read_parquet(OI_PATH)
+        combined = pd.concat([existing, new_row], ignore_index=True)
+    else:
+        OI_PATH.parent.mkdir(parents=True, exist_ok=True)
+        combined = new_row
+
+    combined.to_parquet(OI_PATH, index=False)
+    print(f"[{ts.isoformat()}] OI={oi_value:.2f} → {OI_PATH}")
+
+
+def main():
+    parser = argparse.ArgumentParser(description="OI 장기 수집")
+    parser.add_argument("--symbol", default="XRPUSDT")
+    args = parser.parse_args()
+    collect(symbol=args.symbol)
+
+
+if __name__ == "__main__":
+    main()

scripts/train_model.py

@@ -422,6 +422,113 @@ def walk_forward_auc(
     print(f"  폴드별: {[round(a, 4) for a in aucs]}")
 
 
+def compare(data_path: str, time_weight_decay: float = 2.0, tuned_params_path: str | None = None):
+    """기존 피처 vs OI 파생 피처 추가 버전 A/B 비교."""
+    import warnings
+
+    print("=" * 70)
+    print("  OI 파생 피처 A/B 비교 (30일 데이터 기반, 방향성 참고용)")
+    print("=" * 70)
+
+    df_raw = pd.read_parquet(data_path)
+    base_cols = ["open", "high", "low", "close", "volume"]
+    btc_df = eth_df = None
+    if "close_btc" in df_raw.columns:
+        btc_df = df_raw[[c + "_btc" for c in base_cols]].copy()
+        btc_df.columns = base_cols
+    if "close_eth" in df_raw.columns:
+        eth_df = df_raw[[c + "_eth" for c in base_cols]].copy()
+        eth_df.columns = base_cols
+    df = df_raw[base_cols].copy()
+    if "oi_change" in df_raw.columns:
+        df["oi_change"] = df_raw["oi_change"]
+    if "funding_rate" in df_raw.columns:
+        df["funding_rate"] = df_raw["funding_rate"]
+
+    dataset = generate_dataset_vectorized(
+        df, btc_df=btc_df, eth_df=eth_df,
+        time_weight_decay=time_weight_decay,
+        negative_ratio=5,
+    )
+
+    if dataset.empty:
+        raise ValueError("데이터셋 생성 실패")
+
+    lgbm_params, weight_scale = _load_lgbm_params(tuned_params_path)
+
+    # Baseline: OI 파생 피처 제외
+    BASELINE_EXCLUDE = {"oi_change_ma5", "oi_price_spread"}
+    baseline_cols = [c for c in FEATURE_COLS if c in dataset.columns and c not in BASELINE_EXCLUDE]
+    new_cols = [c for c in FEATURE_COLS if c in dataset.columns]
+
+    results = {}
+    for label, cols in [("Baseline", baseline_cols), ("New", new_cols)]:
+        X = dataset[cols]
+        y = dataset["label"]
+        w = dataset["sample_weight"].values
+        source = dataset["source"].values if "source" in dataset.columns else np.full(len(X), "signal")
+
+        split = int(len(X) * 0.8)
+        X_tr, X_val = X.iloc[:split], X.iloc[split:]
+        y_tr, y_val = y.iloc[:split], y.iloc[split:]
+        w_tr = (w[:split] * weight_scale).astype(np.float32)
+        source_tr = source[:split]
+
+        balanced_idx = stratified_undersample(y_tr.values, source_tr, seed=42)
+        X_tr_b = X_tr.iloc[balanced_idx]
+        y_tr_b = y_tr.iloc[balanced_idx]
+        w_tr_b = w_tr[balanced_idx]
+
+        model = lgb.LGBMClassifier(**lgbm_params, random_state=42, verbose=-1)
+        with warnings.catch_warnings():
+            warnings.simplefilter("ignore")
+            model.fit(X_tr_b, y_tr_b, sample_weight=w_tr_b)
+
+        proba = model.predict_proba(X_val)[:, 1]
+        auc = roc_auc_score(y_val, proba) if len(np.unique(y_val)) > 1 else 0.5
+
+        precs, recs, thrs = precision_recall_curve(y_val, proba)
+        precs, recs = precs[:-1], recs[:-1]
+        valid_idx = np.where(recs >= 0.15)[0]
+        if len(valid_idx) > 0:
+            best_i = valid_idx[np.argmax(precs[valid_idx])]
+            thr, prec, rec = float(thrs[best_i]), float(precs[best_i]), float(recs[best_i])
+        else:
+            thr, prec, rec = 0.50, 0.0, 0.0
+
+        # Feature importance
+        imp = dict(zip(cols, model.feature_importances_))
+        top10 = sorted(imp.items(), key=lambda x: x[1], reverse=True)[:10]
+
+        results[label] = {
+            "auc": auc, "precision": prec, "recall": rec,
+            "threshold": thr, "n_val": len(y_val),
+            "n_val_pos": int(y_val.sum()), "top10": top10,
+        }
+
+    # 비교 테이블 출력
+    n_base = len(baseline_cols)
+    n_new = len(new_cols)
+    print(f"\n{'지표':<20} {f'Baseline({n_base})':>15} {f'New({n_new})':>15} {'Delta':>10}")
+    print("-" * 62)
+    for metric in ["auc", "precision", "recall", "threshold"]:
+        b = results["Baseline"][metric]
+        n = results["New"][metric]
+        d = n - b
+        sign = "+" if d > 0 else ""
+        print(f"{metric:<20} {b:>15.4f} {n:>15.4f} {sign}{d:>9.4f}")
+
+    n_val = results["Baseline"]["n_val"]
+    n_pos = results["Baseline"]["n_val_pos"]
+    print(f"\n검증셋: n={n_val} (양성={n_pos}, 음성={n_val - n_pos})")
+    print("⚠ 30일 데이터 기반 — 방향성 참고용\n")
+
+    print("Feature Importance Top 10 (New):")
+    for feat_name, imp_val in results["New"]["top10"]:
+        marker = " ← NEW" if feat_name in BASELINE_EXCLUDE else ""
+        print(f"  {feat_name:<25} {imp_val:>6}{marker}")
+
+
 def main():
     parser = argparse.ArgumentParser()
     parser.add_argument("--data", default="data/combined_15m.parquet")
@@ -435,9 +542,13 @@ def main():
         "--tuned-params", type=str, default=None,
         help="Optuna 튜닝 결과 JSON 경로 (지정 시 기본 파라미터를 덮어씀)",
     )
+    parser.add_argument("--compare", action="store_true",
+                        help="OI 파생 피처 추가 전후 A/B 성능 비교")
     args = parser.parse_args()
 
-    if args.wf:
+    if args.compare:
+        compare(args.data, time_weight_decay=args.decay, tuned_params_path=args.tuned_params)
+    elif args.wf:
         walk_forward_auc(
             args.data,
             time_weight_decay=args.decay,

src/bot.py

@@ -1,4 +1,5 @@
 import asyncio
+from collections import deque
 import pandas as pd
 from loguru import logger
 from src.config import Config
@@ -24,6 +25,8 @@ class TradingBot:
         self._entry_quantity: float | None = None
         self._is_reentering: bool = False  # _close_and_reenter 중 콜백 상태 초기화 방지
         self._prev_oi: float | None = None  # OI 변화율 계산용 이전 값
+        self._oi_history: deque = deque(maxlen=5)
+        self._latest_ret_1: float = 0.0
         self.stream = MultiSymbolStream(
             symbols=[config.symbol, "BTCUSDT", "ETHUSDT"],
             interval="15m",
@@ -57,21 +60,43 @@ class TradingBot:
         else:
             logger.info("기존 포지션 없음 - 신규 진입 대기")
 
-    async def _fetch_market_microstructure(self) -> tuple[float, float]:
+    async def _init_oi_history(self) -> None:
-        """OI 변화율과 펀딩비를 실시간으로 조회한다. 실패 시 0.0으로 폴백."""
+        """봇 시작 시 최근 OI 변화율 히스토리를 조회하여 deque를 채운다."""
+        try:
+            changes = await self.exchange.get_oi_history(limit=5)
+            for c in changes:
+                self._oi_history.append(c)
+            if changes:
+                self._prev_oi = None
+            logger.info(f"OI 히스토리 초기화: {len(self._oi_history)}개")
+        except Exception as e:
+            logger.warning(f"OI 히스토리 초기화 실패 (무시): {e}")
+
+    async def _fetch_market_microstructure(self) -> tuple[float, float, float, float]:
+        """OI 변화율, 펀딩비, OI MA5, OI-가격 스프레드를 실시간으로 조회한다."""
         oi_val, fr_val = await asyncio.gather(
             self.exchange.get_open_interest(),
             self.exchange.get_funding_rate(),
             return_exceptions=True,
         )
-        # None(API 실패) 또는 Exception이면 _calc_oi_change를 호출하지 않고 0.0 반환
         if isinstance(oi_val, (int, float)) and oi_val > 0:
             oi_change = self._calc_oi_change(float(oi_val))
         else:
             oi_change = 0.0
         fr_float = float(fr_val) if isinstance(fr_val, (int, float)) else 0.0
-        logger.debug(f"OI={oi_val}, OI변화율={oi_change:.6f}, 펀딩비={fr_float:.6f}")
+
-        return oi_change, fr_float
+        # OI 히스토리 업데이트 및 MA5 계산
+        self._oi_history.append(oi_change)
+        oi_ma5 = sum(self._oi_history) / len(self._oi_history) if self._oi_history else 0.0
+
+        # OI-가격 스프레드
+        oi_price_spread = oi_change - self._latest_ret_1
+
+        logger.debug(
+            f"OI={oi_val}, OI변화율={oi_change:.6f}, 펀딩비={fr_float:.6f}, "
+            f"OI_MA5={oi_ma5:.6f}, OI_Price_Spread={oi_price_spread:.6f}"
+        )
+        return oi_change, fr_float, oi_ma5, oi_price_spread
 
     def _calc_oi_change(self, current_oi: float) -> float:
         """이전 OI 대비 변화율을 계산한다. 첫 캔들은 0.0 반환."""
@@ -85,8 +110,14 @@ class TradingBot:
     async def process_candle(self, df, btc_df=None, eth_df=None):
         self.ml_filter.check_and_reload()
 
+        # 가격 수익률 계산 (oi_price_spread용)
+        if len(df) >= 2:
+            prev_close = df["close"].iloc[-2]
+            curr_close = df["close"].iloc[-1]
+            self._latest_ret_1 = (curr_close - prev_close) / prev_close if prev_close != 0 else 0.0
+
         # 캔들 마감 시 OI/펀딩비 실시간 조회 (실패해도 0으로 폴백)
-        oi_change, funding_rate = await self._fetch_market_microstructure()
+        oi_change, funding_rate, oi_ma5, oi_price_spread = await self._fetch_market_microstructure()
 
         if not self.risk.is_trading_allowed():
             logger.warning("리스크 한도 초과 - 거래 중단")
@@ -111,6 +142,7 @@ class TradingBot:
                 df_with_indicators, signal,
                 btc_df=btc_df, eth_df=eth_df,
                 oi_change=oi_change, funding_rate=funding_rate,
+                oi_change_ma5=oi_ma5, oi_price_spread=oi_price_spread,
             )
             if self.ml_filter.is_model_loaded():
                 if not self.ml_filter.should_enter(features):
@@ -126,6 +158,7 @@ class TradingBot:
                     position, raw_signal, df_with_indicators,
                     btc_df=btc_df, eth_df=eth_df,
                     oi_change=oi_change, funding_rate=funding_rate,
+                    oi_change_ma5=oi_ma5, oi_price_spread=oi_price_spread,
                 )
 
     async def _open_position(self, signal: str, df):
@@ -272,6 +305,8 @@ class TradingBot:
         eth_df=None,
         oi_change: float = 0.0,
         funding_rate: float = 0.0,
+        oi_change_ma5: float = 0.0,
+        oi_price_spread: float = 0.0,
     ) -> None:
         """기존 포지션을 청산하고, ML 필터 통과 시 반대 방향으로 즉시 재진입한다."""
         # 재진입 플래그: User Data Stream 콜백이 신규 포지션 상태를 초기화하지 않도록 보호
@@ -288,6 +323,7 @@ class TradingBot:
                     df, signal,
                     btc_df=btc_df, eth_df=eth_df,
                     oi_change=oi_change, funding_rate=funding_rate,
+                    oi_change_ma5=oi_change_ma5, oi_price_spread=oi_price_spread,
                 )
                 if not self.ml_filter.should_enter(features):
                     logger.info(f"ML 필터 차단: {signal} 재진입 무시")
@@ -300,6 +336,7 @@ class TradingBot:
     async def run(self):
         logger.info(f"봇 시작: {self.config.symbol}, 레버리지 {self.config.leverage}x")
         await self._recover_position()
+        await self._init_oi_history()
         balance = await self.exchange.get_balance()
         self.risk.set_base_balance(balance)
         logger.info(f"기준 잔고 설정: {balance:.2f} USDT (동적 증거금 비율 기준점)")

src/dataset_builder.py

@@ -287,8 +287,18 @@ def _calc_features_vectorized(
     else:
         fr_raw = np.full(len(d), np.nan)
 
-    result["oi_change"]    = _rolling_zscore(oi_raw.astype(np.float64))
+    oi_z = _rolling_zscore(oi_raw.astype(np.float64), window=96)
-    result["funding_rate"] = _rolling_zscore(fr_raw.astype(np.float64))
+    result["oi_change"]    = oi_z
+    result["funding_rate"] = _rolling_zscore(fr_raw.astype(np.float64), window=96)
+
+    # --- OI 파생 피처 ---
+    # 1. oi_change_ma5: OI 변화율의 5캔들 이동평균 (단기 추세)
+    oi_series = pd.Series(oi_raw.astype(np.float64))
+    oi_ma5_raw = oi_series.rolling(window=5, min_periods=1).mean().values
+    result["oi_change_ma5"] = _rolling_zscore(oi_ma5_raw, window=96)
+
+    # 2. oi_price_spread: z-scored OI - z-scored 가격 수익률 (연속값)
+    result["oi_price_spread"] = oi_z - ret_1_z
 
     return result
 
@@ -384,7 +394,7 @@ def generate_dataset_vectorized(
     feat_all   = _calc_features_vectorized(d, signal_arr, btc_df=btc_df, eth_df=eth_df)
 
     # 신호 발생 + NaN 없음 + 미래 데이터 충분한 인덱스만
-    OPTIONAL_COLS = {"oi_change", "funding_rate"}
+    OPTIONAL_COLS = {"oi_change", "funding_rate", "oi_change_ma5", "oi_price_spread"}
     available_cols_for_nan_check = [
         c for c in FEATURE_COLS
         if c in feat_all.columns and c not in OPTIONAL_COLS

src/exchange.py

@@ -173,6 +173,30 @@ class BinanceFuturesClient:
             logger.warning(f"펀딩비 조회 실패 (무시): {e}")
             return None
 
+    async def get_oi_history(self, limit: int = 5) -> list[float]:
+        """최근 OI 변화율 히스토리를 조회한다 (봇 초기화용). 실패 시 빈 리스트."""
+        loop = asyncio.get_event_loop()
+        try:
+            result = await loop.run_in_executor(
+                None,
+                lambda: self.client.futures_open_interest_hist(
+                    symbol=self.config.symbol, period="15m", limit=limit + 1,
+                ),
+            )
+            if len(result) < 2:
+                return []
+            oi_values = [float(r["sumOpenInterest"]) for r in result]
+            changes = []
+            for i in range(1, len(oi_values)):
+                if oi_values[i - 1] > 0:
+                    changes.append((oi_values[i] - oi_values[i - 1]) / oi_values[i - 1])
+                else:
+                    changes.append(0.0)
+            return changes
+        except Exception as e:
+            logger.warning(f"OI 히스토리 조회 실패 (무시): {e}")
+            return []
+
     async def create_listen_key(self) -> str:
         """POST /fapi/v1/listenKey — listenKey 신규 발급"""
         loop = asyncio.get_event_loop()

src/ml_features.py

@@ -9,8 +9,9 @@ FEATURE_COLS = [
     "eth_ret_1", "eth_ret_3", "eth_ret_5",
     "xrp_btc_rs", "xrp_eth_rs",
     # 시장 미시구조: OI 변화율(z-score), 펀딩비(z-score)
-    # parquet에 oi_change/funding_rate 컬럼이 없으면 dataset_builder에서 0으로 채움
     "oi_change", "funding_rate",
+    # OI 파생 피처
+    "oi_change_ma5", "oi_price_spread",
     "adx",
 ]
 
@@ -37,12 +38,14 @@ def build_features(
     eth_df: pd.DataFrame | None = None,
     oi_change: float | None = None,
     funding_rate: float | None = None,
+    oi_change_ma5: float | None = None,
+    oi_price_spread: float | None = None,
 ) -> pd.Series:
     """
     기술 지표가 계산된 DataFrame의 마지막 행에서 ML 피처를 추출한다.
-    btc_df, eth_df가 제공되면 24개 피처를, 없으면 16개 피처를 반환한다.
+    btc_df, eth_df가 제공되면 26개 피처를, 없으면 18개 피처를 반환한다.
     signal: "LONG" | "SHORT"
-    oi_change, funding_rate: 실제 값이 제공되면 사용, 없으면 0.0으로 채운다.
+    oi_change, funding_rate, oi_change_ma5, oi_price_spread: 실제 값이 제공되면 사용, 없으면 0.0으로 채운다.
     """
     last = df.iloc[-1]
     close = last["close"]
@@ -132,8 +135,10 @@ def build_features(
         })
 
     # 실시간에서 실제 값이 제공되면 사용, 없으면 0으로 채운다
-    base["oi_change"]    = float(oi_change)    if oi_change    is not None else 0.0
+    base["oi_change"]       = float(oi_change)       if oi_change       is not None else 0.0
-    base["funding_rate"] = float(funding_rate) if funding_rate is not None else 0.0
+    base["funding_rate"]    = float(funding_rate)    if funding_rate    is not None else 0.0
+    base["oi_change_ma5"]   = float(oi_change_ma5)   if oi_change_ma5   is not None else 0.0
+    base["oi_price_spread"] = float(oi_price_spread) if oi_price_spread is not None else 0.0
     base["adx"] = float(last.get("adx", 0))
 
     return pd.Series(base)

tests/test_bot.py

@@ -227,6 +227,42 @@ async def test_process_candle_fetches_oi_and_funding(config, sample_df):
     assert "funding_rate" in call_kwargs
 
 
+def test_bot_has_oi_history_deque(config):
+    """봇이 OI 히스토리 deque를 가져야 한다."""
+    from collections import deque
+    with patch("src.bot.BinanceFuturesClient"):
+        bot = TradingBot(config)
+    assert isinstance(bot._oi_history, deque)
+    assert bot._oi_history.maxlen == 5
+
+
+@pytest.mark.asyncio
+async def test_init_oi_history_fills_deque(config):
+    """_init_oi_history가 deque를 채워야 한다."""
+    with patch("src.bot.BinanceFuturesClient"):
+        bot = TradingBot(config)
+    bot.exchange = AsyncMock()
+    bot.exchange.get_oi_history = AsyncMock(return_value=[0.01, -0.02, 0.03, -0.01, 0.02])
+    await bot._init_oi_history()
+    assert len(bot._oi_history) == 5
+
+
+@pytest.mark.asyncio
+async def test_fetch_microstructure_returns_4_tuple(config):
+    """_fetch_market_microstructure가 4-tuple을 반환해야 한다."""
+    with patch("src.bot.BinanceFuturesClient"):
+        bot = TradingBot(config)
+    bot.exchange = AsyncMock()
+    bot.exchange.get_open_interest = AsyncMock(return_value=5000000.0)
+    bot.exchange.get_funding_rate = AsyncMock(return_value=0.0001)
+    bot._prev_oi = 4900000.0
+    bot._oi_history.extend([0.01, -0.02, 0.03, -0.01])
+    bot._latest_ret_1 = 0.01
+
+    result = await bot._fetch_market_microstructure()
+    assert len(result) == 4
+
+
 def test_calc_oi_change_first_candle_returns_zero(config):
     """첫 캔들은 0.0을 반환하고 _prev_oi를 설정한다."""
     with patch("src.bot.BinanceFuturesClient"):

tests/test_dataset_builder.py

@@ -266,3 +266,74 @@ def test_stratified_undersample_preserves_signal():
     signal_indices = np.where(source == "signal")[0]
     for si in signal_indices:
         assert si in idx, f"signal 인덱스 {si}가 누락됨"
+
+
+def test_oi_derived_features_present():
+    """OI 파생 피처 2개가 결과에 포함되어야 한다."""
+    import numpy as np
+    import pandas as pd
+    from src.dataset_builder import _calc_features_vectorized, _calc_signals, _calc_indicators
+
+    n = 300
+    np.random.seed(42)
+    df = pd.DataFrame({
+        "open":      np.random.uniform(1, 2, n),
+        "high":      np.random.uniform(2, 3, n),
+        "low":       np.random.uniform(0.5, 1, n),
+        "close":     np.random.uniform(1, 2, n),
+        "volume":    np.random.uniform(1000, 5000, n),
+        "oi_change": np.concatenate([np.zeros(100), np.random.uniform(-0.05, 0.05, 200)]),
+    })
+    d = _calc_indicators(df)
+    sig = _calc_signals(d)
+    feat = _calc_features_vectorized(d, sig)
+
+    assert "oi_change_ma5" in feat.columns, "oi_change_ma5 컬럼이 없음"
+    assert "oi_price_spread" in feat.columns, "oi_price_spread 컬럼이 없음"
+
+
+def test_oi_derived_features_nan_when_no_oi():
+    """oi_change 컬럼이 없으면 파생 피처도 nan이어야 한다."""
+    import numpy as np
+    import pandas as pd
+    from src.dataset_builder import _calc_features_vectorized, _calc_signals, _calc_indicators
+
+    n = 200
+    np.random.seed(0)
+    df = pd.DataFrame({
+        "open":   np.random.uniform(1, 2, n),
+        "high":   np.random.uniform(2, 3, n),
+        "low":    np.random.uniform(0.5, 1, n),
+        "close":  np.random.uniform(1, 2, n),
+        "volume": np.random.uniform(1000, 5000, n),
+    })
+    d = _calc_indicators(df)
+    sig = _calc_signals(d)
+    feat = _calc_features_vectorized(d, sig)
+
+    assert feat["oi_change_ma5"].isna().all(), "oi_change 컬럼 없을 때 oi_change_ma5는 전부 nan이어야 함"
+    assert feat["oi_price_spread"].isna().all(), "oi_change 컬럼 없을 때 oi_price_spread는 전부 nan이어야 함"
+
+
+def test_oi_price_spread_is_continuous():
+    """oi_price_spread는 바이너리가 아닌 연속값이어야 한다."""
+    import numpy as np
+    import pandas as pd
+    from src.dataset_builder import _calc_features_vectorized, _calc_signals, _calc_indicators
+
+    n = 300
+    np.random.seed(42)
+    df = pd.DataFrame({
+        "open":      np.random.uniform(1, 2, n),
+        "high":      np.random.uniform(2, 3, n),
+        "low":       np.random.uniform(0.5, 1, n),
+        "close":     np.random.uniform(1, 2, n),
+        "volume":    np.random.uniform(1000, 5000, n),
+        "oi_change": np.random.uniform(-0.05, 0.05, n),
+    })
+    d = _calc_indicators(df)
+    sig = _calc_signals(d)
+    feat = _calc_features_vectorized(d, sig)
+
+    valid = feat["oi_price_spread"].dropna()
+    assert len(valid.unique()) > 2, "oi_price_spread는 연속값이어야 함 (2개 초과 유니크 값)"

tests/test_exchange.py

@@ -113,3 +113,43 @@ async def test_get_funding_rate_error_returns_none(exchange):
     )
     result = await exchange.get_funding_rate()
     assert result is None
+
+
+@pytest.mark.asyncio
+async def test_get_oi_history_returns_changes(exchange):
+    """get_oi_history()가 OI 변화율 리스트를 반환하는지 확인."""
+    exchange.client.futures_open_interest_hist = MagicMock(
+        return_value=[
+            {"sumOpenInterest": "1000000"},
+            {"sumOpenInterest": "1010000"},
+            {"sumOpenInterest": "1005000"},
+            {"sumOpenInterest": "1020000"},
+            {"sumOpenInterest": "1015000"},
+            {"sumOpenInterest": "1030000"},
+        ]
+    )
+    result = await exchange.get_oi_history(limit=5)
+    assert len(result) == 5
+    assert isinstance(result[0], float)
+    # 첫 번째 변화율: (1010000 - 1000000) / 1000000 = 0.01
+    assert abs(result[0] - 0.01) < 1e-6
+
+
+@pytest.mark.asyncio
+async def test_get_oi_history_error_returns_empty(exchange):
+    """API 오류 시 빈 리스트 반환 확인."""
+    exchange.client.futures_open_interest_hist = MagicMock(
+        side_effect=Exception("API error")
+    )
+    result = await exchange.get_oi_history(limit=5)
+    assert result == []
+
+
+@pytest.mark.asyncio
+async def test_get_oi_history_insufficient_data_returns_empty(exchange):
+    """데이터가 부족하면 빈 리스트 반환 확인."""
+    exchange.client.futures_open_interest_hist = MagicMock(
+        return_value=[{"sumOpenInterest": "1000000"}]
+    )
+    result = await exchange.get_oi_history(limit=5)
+    assert result == []

tests/test_ml_features.py

@@ -21,17 +21,17 @@ def _make_df(n=10, base_price=1.0):
     })
 
 
-def test_build_features_with_btc_eth_has_24_features():
+def test_build_features_with_btc_eth_has_26_features():
     xrp_df = _make_df(10, base_price=1.0)
     btc_df = _make_df(10, base_price=50000.0)
     eth_df = _make_df(10, base_price=3000.0)
     features = build_features(xrp_df, "LONG", btc_df=btc_df, eth_df=eth_df)
-    assert len(features) == 24
+    assert len(features) == 26
 
-def test_build_features_without_btc_eth_has_16_features():
+def test_build_features_without_btc_eth_has_18_features():
     xrp_df = _make_df(10, base_price=1.0)
     features = build_features(xrp_df, "LONG")
-    assert len(features) == 16
+    assert len(features) == 18
 
 def test_build_features_btc_ret_1_correct():
     xrp_df = _make_df(10, base_price=1.0)
@@ -51,8 +51,9 @@ def test_build_features_rs_zero_when_btc_ret_zero():
     assert features["xrp_btc_rs"] == 0.0
 
 def test_feature_cols_has_24_items():
+    """Legacy test — updated to 26 after OI derived features added."""
     from src.ml_features import FEATURE_COLS
-    assert len(FEATURE_COLS) == 24
+    assert len(FEATURE_COLS) == 26
 
 
 def make_df(n=100):
@@ -139,3 +140,31 @@ def test_build_features_defaults_to_zero_when_not_provided(sample_df_with_indica
     feat = build_features(sample_df_with_indicators, signal="LONG")
     assert feat["oi_change"] == pytest.approx(0.0)
     assert feat["funding_rate"] == pytest.approx(0.0)
+
+
+def test_feature_cols_has_26_items():
+    from src.ml_features import FEATURE_COLS
+    assert len(FEATURE_COLS) == 26
+
+
+def test_build_features_with_oi_derived_params():
+    """oi_change_ma5, oi_price_spread 파라미터가 피처에 반영된다."""
+    xrp_df = _make_df(10, base_price=1.0)
+    btc_df = _make_df(10, base_price=50000.0)
+    eth_df = _make_df(10, base_price=3000.0)
+    features = build_features(
+        xrp_df, "LONG",
+        btc_df=btc_df, eth_df=eth_df,
+        oi_change=0.05, funding_rate=0.0002,
+        oi_change_ma5=0.03, oi_price_spread=0.12,
+    )
+    assert features["oi_change_ma5"] == pytest.approx(0.03)
+    assert features["oi_price_spread"] == pytest.approx(0.12)
+
+
+def test_build_features_oi_derived_defaults_to_zero():
+    """oi_change_ma5, oi_price_spread 미제공 시 0.0으로 채워진다."""
+    xrp_df = _make_df(10, base_price=1.0)
+    features = build_features(xrp_df, "LONG")
+    assert features["oi_change_ma5"] == pytest.approx(0.0)
+    assert features["oi_price_spread"] == pytest.approx(0.0)