### 成本敏感半监督集成算法实现 以下是基于 Python 的成本敏感半监督集成学习方法的代码示例，适用于处理不平衡数据集中的二分类问题（如反欺诈）。该代码实现了自训练法(Self-training)，并结合了成本敏感机制来优化模型性能。 #### 自定义成本敏感半监督集成算法 ```python import numpy as np from sklearn.semi_supervised import SelfTrainingClassifier from sklearn.linear_model import LogisticRegression from sklearn.metrics import precision_score, recall_score, f1_score, roc_auc_score, make_scorer, geometric_mean_score from sklearn.model_selection import train_test_split from imblearn.over_sampling import SMOTE from collections import Counter def cost_sensitive_self_training(X_labeled, y_labeled, X_unlabeled, base_estimator=None): """ Cost-Sensitive Semi-Supervised Ensemble Learning Algorithm. Parameters: X_labeled: Labeled feature matrix. y_labeled: Labels corresponding to labeled features. X_unlabeled: Unlabeled feature matrix. base_estimator: Base classifier used within self-training. Returns: Trained model and predictions on unlabeled data. """ if base_estimator is None: base_estimator = LogisticRegression(class_weight='balanced', max_iter=1000) # Apply SMOTE to handle imbalance in the labeled dataset smote = SMOTE(random_state=42) X_resampled, y_resampled = smote.fit_resample(X_labeled, y_labeled) # Initialize a self-training classifier with the resampled data st_clf = SelfTrainingClassifier(base_estimator, threshold=0.75, verbose=True) st_clf.fit(np.vstack((X_resampled, X_unlabeled)), list(y_resampled) + [np.nan]*len(X_unlabeled)) return st_clf # Evaluation Metrics Function def evaluate_metrics(y_true, y_pred_proba, thresholds=[0.5]): results = [] for t in thresholds: y_pred = (y_pred_proba >= t).astype(int) prec = precision_score(y_true, y_pred) rec = recall_score(y_true, y_pred) f1 = f1_score(y_true, y_pred) gmean = geometric_mean_score(y_true, y_pred) auc = roc_auc_score(y_true, y_pred_proba) metrics_dict = { 'Threshold': t, 'Precision': prec, 'Recall': rec, 'F1 Score': f1, 'G-Mean': gmean, 'AUC': auc } results.append(metrics_dict) return results # Example Usage if __name__ == "__main__": from sklearn.datasets import make_classification # Generate synthetic imbalanced dataset X, y = make_classification(n_samples=1000, n_features=20, weights=[0.95], flip_y=0, random_state=42) X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3, stratify=y, random_state=42) # Split into labeled and unlabeled datasets X_labeled, X_unlabeled, y_labeled, _ = train_test_split( X_train, y_train, test_size=0.8, stratify=y_train, random_state=42 ) clf = cost_sensitive_self_training(X_labeled, y_labeled, X_unlabeled) y_pred_proba = clf.predict_proba(X_test)[:, 1] eval_results = evaluate_metrics(y_test, y_pred_proba) for result in eval_results: print(result) ``` 上述代码通过 `SelfTrainingClassifier` 实现了一个简单的自训练框架，并引入了合成少数类过采样技术(SMOTE)[^1] 来解决类别不均衡问题。此外，在评估阶段计算 Precision、Recall、F1 分数、几何均值(G-mean) 和 AUC 值以全面衡量模型表现。 --- ### 关键概念解释 - **Cost-Sensitivity**: 在机器学习中，当错误代价不对称时，可以通过调整权重或阈值使模型更关注于减少高代价误分类的情况[^1]。 - **Semi-Supervised Learning**: 利用少量标注样本和大量未标注样本进行建模的方法可以有效缓解标签稀缺的问题。 - **Evaluation Metrics**: 使用多种评价标准能够更好地理解模型在不同场景下的适用性和鲁棒性[^1]。 ---