因為在金融數據預測裡，時間序列的資料不能隨機打亂，否則會發生「未來資訊洩漏到過去」的情況，導致模型表現看似很好，但實際上完全不符合交易場景，所以將原來的ML訓練模型改成Walk Forward Validation 版本，跟之前的差別在於原來的只有一次切分，調整後有多次切分

def train_ml_model_walkforward(X, y, n_splits=5):
    """
    使用 Walk Forward Validation 訓練模型
    :param X: 特徵資料
    :param y: 標籤
    :param n_splits: 分割次數 (k-fold)
    """
    tscv = TimeSeriesSplit(n_splits=n_splits)
    all_scores = []
    split_num = 1

    for train_index, test_index in tscv.split(X):
        X_train, X_test = X.iloc[train_index], X.iloc[test_index]
        y_train, y_test = y.iloc[train_index], y.iloc[test_index]

        # 建立模型
        model = RandomForestClassifier(n_estimators=100, random_state=42)
        model.fit(X_train, y_train)
        y_pred = model.predict(X_test)

        # 評估
        acc = accuracy_score(y_test, y_pred)
        all_scores.append(acc)

        print(f"=== Split {split_num} ===")
        print(f"Train size: {len(train_index)}, Test size: {len(test_index)}")
        print(f"Accuracy: {acc:.4f}")
        print("Confusion Matrix:")
        print(confusion_matrix(y_test, y_pred))
        print("Classification Report:")
        print(classification_report(y_test, y_pred))
        print("\n")
        
        split_num += 1

    print("平均準確率:", sum(all_scores) / len(all_scores))
    return all_scores
# -------------------------
# 主程式
# -------------------------
if __name__ == "__main__":
    df = fetch_crypto_data("BTC/USDT", "1h", 500)
    df = add_indicators(df)

    X, y = prepare_ml_data(df)
    scores = train_ml_model_walkforward(X, y, n_splits=5)