\u8be5\u811a\u672c\u5c06\u5305\u542b\u4ee5\u4e0b\u529f\u80fd\uff1a\u4f7f\u7528\u968f\u673a\u68ee\u6797\uff08Random Forest\uff09\u8fdb\u884c\u6709\u76d1\u7763\u7684\u6b3a\u8bc8\u68c0\u6d4b<\/p>\n\n\n\n
<\/p>\n\n\n\n
import pandas as pd\nimport numpy as np\nfrom sklearn.model_selection import train_test_split\nfrom sklearn.ensemble import IsolationForest, RandomForestClassifier\nfrom sklearn.neighbors import LocalOutlierFactor\nfrom sklearn.metrics import accuracy_score, precision_score, recall_score, f1_score, roc_auc_score, confusion_matrix, classification_report\nfrom sklearn.preprocessing import StandardScaler, LabelEncoder\nimport matplotlib.pyplot as plt\nimport seaborn as sns\nfrom datetime import datetime, timedelta\nimport warnings\nwarnings.filterwarnings(\"ignore\")\n\n# --- \u914d\u7f6e ---\nNUM_TRANSACTIONS = 10000\nFRAUD_RATIO = 0.05 # 5% \u7684\u6b3a\u8bc8\u4ea4\u6613\nREPORT_PREFIX = '\u7535\u5546\u6b3a\u8bc8\u68c0\u6d4b\u62a5\u544a'\nRANDOM_SEED = 42\n\n# --- \u6570\u636e\u751f\u6210 ---\n\ndef generate_sample_fraud_data(n_transactions, fraud_ratio):\n \"\"\"\u751f\u6210\u6a21\u62df\u7684\u7535\u5546\u4ea4\u6613\u6570\u636e\uff0c\u5305\u542b\u6b63\u5e38\u548c\u6b3a\u8bc8\u4ea4\u6613\"\"\"\n print(\"--- \u6b63\u5728\u751f\u6210\u6a21\u62df\u4ea4\u6613\u6570\u636e ---\")\n np.random.seed(RANDOM_SEED)\n \n data = []\n transaction_ids = [f'trans_{i}' for i in range(1, n_transactions + 1)]\n \n n_fraud = int(n_transactions * fraud_ratio)\n n_normal = n_transactions - n_fraud\n print(f\"\u751f\u6210 {n_normal} \u7b14\u6b63\u5e38\u4ea4\u6613 \u548c {n_fraud} \u7b14\u6b3a\u8bc8\u4ea4\u6613...\")\n \n for i in range(n_transactions):\n is_fraud = 1 if i < n_fraud else 0\n \n # --- \u57fa\u7840\u4ea4\u6613\u4fe1\u606f ---\n user_id = f'user_{np.random.randint(1, n_transactions \/\/ 10)}' # \u5047\u8bbe\u7528\u6237\u6570\u662f\u4ea4\u6613\u6570\u76841\/10\n product_id = f'prod_{np.random.randint(1, 1000)}'\n amount = np.random.lognormal(7, 1.5) # \u4ea4\u6613\u91d1\u989d (\u5bf9\u6570\u6b63\u6001\u5206\u5e03)\n timestamp = datetime.now() - timedelta(days=np.random.randint(0, 365), seconds=np.random.randint(0, 86400))\n \n # --- \u7528\u6237\u884c\u4e3a\u7279\u5f81 ---\n # \u7528\u6237\u5386\u53f2\u4ea4\u6613\u6b21\u6570\u548c\u603b\u91d1\u989d\n user_tx_count = np.random.poisson(20) if not is_fraud else np.random.poisson(5)\n user_total_spent = np.random.lognormal(9, 1.0) if not is_fraud else np.random.lognormal(7, 1.5)\n \n # \u7528\u6237\u5e73\u5747\u4ea4\u6613\u91d1\u989d\n user_avg_amount = user_total_spent \/ max(user_tx_count, 1)\n \n # --- \u8bbe\u5907\u548c\u5730\u7406\u4f4d\u7f6e\u7279\u5f81 ---\n device_id = f'device_{np.random.randint(1, 5000)}'\n ip_address = f\"192.168.{np.random.randint(0, 255)}.{np.random.randint(0, 255)}\"\n \n # \u5730\u7406\u4f4d\u7f6e (\u7b80\u5316\u4e3a\u56fd\u5bb6\u548c\u57ce\u5e02ID)\n country = np.random.choice(['CN', 'US', 'UK', 'JP'], p=[0.6, 0.2, 0.1, 0.1])\n city_id = np.random.randint(1, 100)\n \n # --- \u6b3a\u8bc8\u6a21\u5f0f\u6a21\u62df ---\n if is_fraud:\n # \u6b3a\u8bc8\u4ea4\u6613\u53ef\u80fd\u5177\u6709\u7684\u7279\u5f81\n # 1. \u91d1\u989d\u5f02\u5e38\u9ad8\n if np.random.random() < 0.3:\n amount *= np.random.uniform(2, 10)\n \n # 2. \u77ed\u65f6\u95f4\u5185\u591a\u6b21\u4ea4\u6613 (Velocity)\n if np.random.random() < 0.4:\n # \u8fd9\u91cc\u7b80\u5316\u5904\u7406\uff0c\u5728\u6570\u636e\u751f\u6210\u65f6\u65e0\u6cd5\u5b8c\u5168\u6a21\u62df\u65f6\u95f4\u5e8f\u5217\uff0c\u4f46\u5728\u7279\u5f81\u5de5\u7a0b\u4e2d\u53ef\u4ee5\u4f53\u73b0\n pass \n \n # 3. \u65b0\u8bbe\u5907\/\u65b0IP\n if np.random.random() < 0.5:\n device_id = f'device_fraud_{np.random.randint(10000, 20000)}' # \u4f7f\u7528\u4e00\u4e2a\u770b\u8d77\u6765\u50cf\u65b0\u8bbe\u5907\u7684ID\n ip_address = f\"10.{np.random.randint(0, 255)}.{np.random.randint(0, 255)}.{np.random.randint(0, 255)}\"\n\n # 4. \u8de8\u5730\u533a\u4ea4\u6613 (\u7528\u6237\u5e38\u7528\u5730\u533a vs \u672c\u6b21\u4ea4\u6613\u5730\u533a)\n # \u7b80\u5316\uff1a\u5047\u8bbe\u6b3a\u8bc8\u4ea4\u6613\u66f4\u53ef\u80fd\u6765\u81ea\u4e0d\u540c\u56fd\u5bb6\n if np.random.random() < 0.4:\n country = np.random.choice(['RU', 'BR', 'IN']) # \u6a21\u62df\u9ad8\u98ce\u9669\u56fd\u5bb6\n \n # 5. \u7528\u6237\u884c\u4e3a\u5f02\u5e38 (\u4f8b\u5982\uff0c\u65b0\u7528\u6237\u5927\u989d\u4ea4\u6613)\n if np.random.random() < 0.3:\n user_tx_count = np.random.poisson(1) # \u65b0\u7528\u6237\n user_total_spent = amount # \u603b\u91d1\u989d\u5c31\u662f\u672c\u6b21\u4ea4\u6613\u91d1\u989d\n\n data.append({\n 'transaction_id': transaction_ids[i],\n 'user_id': user_id,\n 'product_id': product_id,\n 'amount': round(amount, 2),\n 'timestamp': timestamp,\n 'user_tx_count': user_tx_count,\n 'user_total_spent': round(user_total_spent, 2),\n 'user_avg_amount': round(user_avg_amount, 2),\n 'device_id': device_id,\n 'ip_address': ip_address,\n 'country': country,\n 'city_id': city_id,\n 'is_fraud': is_fraud\n })\n \n df = pd.DataFrame(data)\n \n # --- \u884d\u751f\u7279\u5f81\u5de5\u7a0b (\u5728\u6570\u636e\u751f\u6210\u540e\u8fdb\u884c) ---\n print(\"\u6b63\u5728\u8fdb\u884c\u7279\u5f81\u5de5\u7a0b...\")\n df['timestamp'] = pd.to_datetime(df['timestamp'])\n \n # 1. \u65f6\u95f4\u7279\u5f81\n df['hour_of_day'] = df['timestamp'].dt.hour\n df['day_of_week'] = df['timestamp'].dt.dayofweek\n \n # 2. \u7528\u6237\u884c\u4e3a\u504f\u5dee\u7279\u5f81\n df['amount_to_user_avg_ratio'] = df['amount'] \/ (df['user_avg_amount'] + 1e-8) # \u907f\u514d\u9664\u4ee5\u96f6\n \n # 3. \u901f\u5ea6\u7279\u5f81 (\u6bcf\u4e2a\u7528\u6237\u6bcf\u5c0f\u65f6\u4ea4\u6613\u6b21\u6570) - \u7b80\u5316\u6a21\u62df\n user_hourly_counts = df.groupby(['user_id', pd.Grouper(key='timestamp', freq='H')]).size().reset_index(name='tx_per_hour')\n df = df.merge(user_hourly_counts, on=['user_id', 'timestamp'], how='left')\n df['tx_per_hour'].fillna(0, inplace=True)\n \n # 4. \u8bbe\u5907\/IP\u65b0\u9c9c\u5ea6 (\u5047\u8bbe\u6709\u4e00\u4e2a\u8bb0\u5f55\u9996\u6b21\u51fa\u73b0\u65f6\u95f4\u7684\u8868\uff0c\u8fd9\u91cc\u7b80\u5316\u4e3a\u662f\u5426\u662f\u65b0\u8bbe\u5907)\n # \u7b80\u5316\u5904\u7406\uff1a\u5982\u679cdevice_id\u5305\u542b'fraud'\u5219\u8ba4\u4e3a\u662f\u65b0\u8bbe\u5907\n df['is_new_device'] = df['device_id'].str.contains('fraud').astype(int)\n \n csv_filename = f'{REPORT_PREFIX}_\u6a21\u62df\u6570\u636e.csv'\n df.to_csv(csv_filename, index=False, encoding='utf-8-sig')\n print(f\"\u6a21\u62df\u6570\u636e\u5df2\u751f\u6210\u5e76\u4fdd\u5b58\u81f3: {csv_filename}\")\n return df\n\n# --- \u6570\u636e\u9884\u5904\u7406 ---\n\ndef preprocess_data(df):\n \"\"\"\u6570\u636e\u9884\u5904\u7406\"\"\"\n print(\"\\n--- \u6b63\u5728\u8fdb\u884c\u6570\u636e\u9884\u5904\u7406 ---\")\n df_processed = df.copy()\n \n # 1. \u7f16\u7801\u5206\u7c7b\u53d8\u91cf\n le_country = LabelEncoder()\n df_processed['country_encoded'] = le_country.fit_transform(df_processed['country'])\n \n # 2. \u9009\u62e9\u7528\u4e8e\u5efa\u6a21\u7684\u7279\u5f81\u5217 (\u6392\u9664ID\u548c\u6807\u7b7e)\n feature_columns = [\n 'amount', 'user_tx_count', 'user_total_spent', 'user_avg_amount',\n 'country_encoded', 'city_id', 'hour_of_day', 'day_of_week',\n 'amount_to_user_avg_ratio', 'tx_per_hour', 'is_new_device'\n ]\n \n X = df_processed[feature_columns]\n y = df_processed['is_fraud']\n \n print(f\"\u9884\u5904\u7406\u5b8c\u6210\u3002\u7279\u5f81\u77e9\u9635\u5f62\u72b6: {X.shape}, \u6807\u7b7e\u5411\u91cf\u5f62\u72b6: {y.shape}\")\n # \u6253\u5370\u7c7b\u522b\u5206\u5e03\n print(f\"\u6b3a\u8bc8\u4ea4\u6613\u6570\u91cf: {y.sum()}, \u5360\u6bd4: {y.mean():.2%}\")\n return X, y, le_country, feature_columns\n\n# --- \u65e0\u76d1\u7763\u5f02\u5e38\u68c0\u6d4b ---\n\ndef perform_unsupervised_detection(X, y):\n \"\"\"\u6267\u884c\u65e0\u76d1\u7763\u5f02\u5e38\u68c0\u6d4b\"\"\"\n print(\"\\n--- \u6b63\u5728\u6267\u884c\u65e0\u76d1\u7763\u5f02\u5e38\u68c0\u6d4b ---\")\n results = {}\n \n # --- 1. Isolation Forest ---\n print(\"\u8bad\u7ec3 Isolation Forest \u6a21\u578b...\")\n # \u6ce8\u610f\uff1aIsolationForest\u9884\u6d4b1\u4e3a\u6b63\u5e38\uff0c-1\u4e3a\u5f02\u5e38\uff0c\u9700\u8981\u8f6c\u6362\n iso_forest = IsolationForest(contamination=0.05, random_state=RANDOM_SEED, n_estimators=100)\n iso_forest.fit(X)\n y_pred_iso = iso_forest.predict(X)\n y_pred_iso = np.where(y_pred_iso == -1, 1, 0) # \u8f6c\u6362\u4e3a 1 (\u6b3a\u8bc8), 0 (\u6b63\u5e38)\n \n results['Isolation Forest'] = {\n 'predictions': y_pred_iso,\n 'accuracy': accuracy_score(y, y_pred_iso),\n 'precision': precision_score(y, y_pred_iso, zero_division=0),\n 'recall': recall_score(y, y_pred_iso),\n 'f1': f1_score(y, y_pred_iso),\n 'auc': roc_auc_score(y, y_pred_iso) if len(np.unique(y_pred_iso)) > 1 else 0.0\n }\n \n # --- 2. Local Outlier Factor ---\n print(\"\u8bad\u7ec3 Local Outlier Factor \u6a21\u578b...\")\n # LOF\u4e0d\u9700\u8981\u663e\u5f0f\u8bad\u7ec3\uff0c\u76f4\u63a5fit_predict\n lof = LocalOutlierFactor(n_neighbors=20, contamination=0.05)\n y_pred_lof = lof.fit_predict(X)\n y_pred_lof_scores = -lof.negative_outlier_factor_ # \u83b7\u53d6\u5f02\u5e38\u8bc4\u5206\n y_pred_lof = np.where(y_pred_lof == -1, 1, 0) # \u8f6c\u6362\u4e3a 1 (\u6b3a\u8bc8), 0 (\u6b63\u5e38)\n \n results['Local Outlier Factor'] = {\n 'predictions': y_pred_lof,\n 'scores': y_pred_lof_scores, # \u4fdd\u7559\u8bc4\u5206\u7528\u4e8e\u53ef\u80fd\u7684\u9608\u503c\u8c03\u6574\n 'accuracy': accuracy_score(y, y_pred_lof),\n 'precision': precision_score(y, y_pred_lof, zero_division=0),\n 'recall': recall_score(y, y_pred_lof),\n 'f1': f1_score(y, y_pred_lof),\n 'auc': roc_auc_score(y, y_pred_lof) if len(np.unique(y_pred_lof)) > 1 else 0.0\n }\n \n # --- \u62a5\u544a ---\n print(\"\\n--- \u65e0\u76d1\u7763\u6a21\u578b\u6027\u80fd\u5bf9\u6bd4 ---\")\n unsup_comparison_df = pd.DataFrame({\n 'Model': list(results.keys()),\n 'Accuracy': [results[k]['accuracy'] for k in results.keys()],\n 'Precision': [results[k]['precision'] for k in results.keys()],\n 'Recall': [results[k]['recall'] for k in results.keys()],\n 'F1-Score': [results[k]['f1'] for k in results.keys()],\n 'AUC-ROC': [results[k]['auc'] for k in results.keys()]\n })\n print(unsup_comparison_df.round(4).to_string(index=False))\n \n return results\n\n# --- \u6709\u76d1\u7763\u6b3a\u8bc8\u68c0\u6d4b ---\n\ndef perform_supervised_detection(X, y):\n \"\"\"\u6267\u884c\u6709\u76d1\u7763\u6b3a\u8bc8\u68c0\u6d4b\"\"\"\n print(\"\\n--- \u6b63\u5728\u6267\u884c\u6709\u76d1\u7763\u6b3a\u8bc8\u68c0\u6d4b ---\")\n \n # \u5212\u5206\u8bad\u7ec3\u96c6\u548c\u6d4b\u8bd5\u96c6 (\u6a21\u62df\u5728\u7ebf\u5b66\u4e60\uff0c\u7528\u5386\u53f2\u6570\u636e\u8bad\u7ec3\uff0c\u65b0\u6570\u636e\u6d4b\u8bd5)\n # \u4e3a\u4e86\u6f14\u793a\uff0c\u6211\u4eec\u7b80\u5355\u5730\u5212\u5206\u6570\u636e\u96c6\n X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3, random_state=RANDOM_SEED, stratify=y)\n \n # \u7279\u5f81\u7f29\u653e\n scaler = StandardScaler()\n X_train_scaled = scaler.fit_transform(X_train)\n X_test_scaled = scaler.transform(X_test)\n \n # --- Random Forest Classifier ---\n print(\"\u8bad\u7ec3 Random Forest \u6a21\u578b...\")\n model_rf = RandomForestClassifier(n_estimators=100, random_state=RANDOM_SEED, class_weight='balanced', n_jobs=-1) # \u5904\u7406\u7c7b\u522b\u4e0d\u5e73\u8861\n model_rf.fit(X_train_scaled, y_train)\n y_pred_rf = model_rf.predict(X_test_scaled)\n y_pred_proba_rf = model_rf.predict_proba(X_test_scaled)[:, 1]\n \n results = {\n 'Random Forest': {\n 'model': model_rf,\n 'scaler': scaler,\n 'predictions': y_pred_rf,\n 'probabilities': y_pred_proba_rf,\n 'accuracy': accuracy_score(y_test, y_pred_rf),\n 'precision': precision_score(y_test, y_pred_rf),\n 'recall': recall_score(y_test, y_pred_rf),\n 'f1': f1_score(y_test, y_pred_rf),\n 'auc': roc_auc_score(y_test, y_pred_proba_rf),\n 'y_test': y_test # \u4fdd\u5b58\u6d4b\u8bd5\u6807\u7b7e\u7528\u4e8e\u62a5\u544a\n }\n }\n \n # --- \u62a5\u544a ---\n rf_metrics = results['Random Forest']\n print(f\"\\n--- Random Forest \u6a21\u578b\u6027\u80fd ---\")\n print(f\"Accuracy: {rf_metrics['accuracy']:.4f}\")\n print(f\"Precision: {rf_metrics['precision']:.4f}\")\n print(f\"Recall: {rf_metrics['recall']:.4f}\")\n print(f\"F1-Score: {rf_metrics['f1']:.4f}\")\n print(f\"AUC-ROC: {rf_metrics['auc']:.4f}\")\n \n # \u7ed8\u5236\u6df7\u6dc6\u77e9\u9635\n cm = confusion_matrix(rf_metrics['y_test'], rf_metrics['predictions'])\n plt.figure(figsize=(8, 6))\n sns.heatmap(cm, annot=True, fmt='d', cmap='Reds', \n xticklabels=['Normal', 'Fraud'], \n yticklabels=['Normal', 'Fraud'])\n plt.title('\u6df7\u6dc6\u77e9\u9635 - Random Forest')\n plt.xlabel('\u9884\u6d4b\u6807\u7b7e')\n plt.ylabel('\u771f\u5b9e\u6807\u7b7e')\n cm_path = f'{REPORT_PREFIX}_\u6df7\u6dc6\u77e9\u9635_RandomForest.png'\n plt.savefig(cm_path)\n plt.close()\n print(f\"\u6df7\u6dc6\u77e9\u9635\u56fe\u8868\u5df2\u4fdd\u5b58\u81f3: {cm_path}\")\n \n # \u6253\u5370\u8be6\u7ec6\u5206\u7c7b\u62a5\u544a\n print(\"\\n--- Random Forest \u8be6\u7ec6\u5206\u7c7b\u62a5\u544a ---\")\n print(classification_report(rf_metrics['y_test'], rf_metrics['predictions'], target_names=['Normal', 'Fraud']))\n \n return results, cm_path\n\n# --- \u7279\u5f81\u91cd\u8981\u6027\u5206\u6790 ---\n\ndef analyze_feature_importance(model, feature_names, model_name):\n \"\"\"\u5206\u6790\u5e76\u53ef\u89c6\u5316\u7279\u5f81\u91cd\u8981\u6027\"\"\"\n print(f\"\\n--- \u5206\u6790 {model_name} \u7279\u5f81\u91cd\u8981\u6027 ---\")\n \n if hasattr(model, 'feature_importances_'):\n importances = model.feature_importances_\n indices = np.argsort(importances)[::-1]\n title = f'{model_name} - \u7279\u5f81\u91cd\u8981\u6027'\n else:\n print(\"\u6a21\u578b\u4e0d\u652f\u6301\u7279\u5f81\u91cd\u8981\u6027\u5206\u6790\u3002\")\n return None, None\n\n # \u521b\u5efa\u7279\u5f81\u91cd\u8981\u6027DataFrame\n feature_importance_df = pd.DataFrame({\n 'feature': [feature_names[i] for i in indices],\n 'importance': [importances[i] for i in indices]\n })\n \n print(\"\u7279\u5f81\u91cd\u8981\u6027\u6392\u5e8f:\")\n print(feature_importance_df.head(10).to_string(index=False))\n\n # \u7ed8\u5236\u7279\u5f81\u91cd\u8981\u6027\n plt.figure(figsize=(10, 6))\n sns.barplot(data=feature_importance_df.head(10), x='importance', y='feature', palette='viridis')\n plt.title(title)\n plt.xlabel('\u91cd\u8981\u6027')\n plt.tight_layout()\n feat_imp_path = f'{REPORT_PREFIX}_\u7279\u5f81\u91cd\u8981\u6027_{model_name.replace(\" \", \"_\")}.png'\n plt.savefig(feat_imp_path)\n plt.close()\n print(f\"\u7279\u5f81\u91cd\u8981\u6027\u56fe\u8868\u5df2\u4fdd\u5b58\u81f3: {feat_imp_path}\")\n \n return feature_importance_df, feat_imp_path\n\n# --- \u62a5\u544a\u751f\u6210 ---\n\ndef generate_fraud_detection_report(unsup_results, sup_results, cm_path, feat_imp_df, feat_imp_path):\n \"\"\"\u751f\u6210\u6700\u7ec8\u7684\u6b3a\u8bc8\u68c0\u6d4b\u5206\u6790\u62a5\u544a\"\"\"\n print(\"\\n--- \u6b63\u5728\u751f\u6210\u6b3a\u8bc8\u68c0\u6d4b\u5206\u6790\u62a5\u544a ---\")\n from datetime import datetime\n report_filename = f\"{REPORT_PREFIX}_{datetime.now().strftime('%Y%m%d_%H%M%S')}.txt\"\n \n sup_metrics = sup_results['Random Forest']\n \n with open(report_filename, 'w', encoding='utf-8') as f:\n f.write(\"=\" * 50 + \"\\n\")\n f.write(\" \u7535\u5546\u5e73\u53f0\u7528\u6237\u98ce\u9669\u7ba1\u7406\u548c\u6b3a\u8bc8\u884c\u4e3a\u68c0\u6d4b\u62a5\u544a\\n\")\n f.write(f\" \u751f\u6210\u65f6\u95f4: {datetime.now().strftime('%Y-%m-%d %H:%M:%S')}\\n\")\n f.write(\"=\" * 50 + \"\\n\\n\")\n\n f.write(\"--- 1. \u9879\u76ee\u6982\u8ff0 ---\\n\")\n f.write(\"\u672c\u9879\u76ee\u65e8\u5728\u901a\u8fc7\u5206\u6790\u7535\u5546\u4ea4\u6613\u6570\u636e\uff0c\u8fd0\u7528\u673a\u5668\u5b66\u4e60\u6280\u672f\u8bc6\u522b\u6f5c\u5728\u7684\u6b3a\u8bc8\u884c\u4e3a\uff0c\u4ee5\u4fdd\u62a4\u5e73\u53f0\u548c\u7528\u6237\u7684\u8d44\u91d1\u5b89\u5168\u3002\\n\")\n f.write(\"\u91c7\u7528\u4e86\u65e0\u76d1\u7763\u5b66\u4e60\uff08\u5f02\u5e38\u68c0\u6d4b\uff09\u548c\u6709\u76d1\u7763\u5b66\u4e60\uff08\u5206\u7c7b\uff09\u4e24\u79cd\u65b9\u6cd5\u8fdb\u884c\u5bf9\u6bd4\u5206\u6790\u3002\\n\\n\")\n\n f.write(\"--- 2. \u6570\u636e\u6982\u89c8 ---\\n\")\n f.write(\"\u6570\u636e\u6765\u6e90: \u6a21\u62df\u751f\u6210\u7684\u7535\u5546\u5e73\u53f0\u4ea4\u6613\u6570\u636e\u3002\\n\")\n f.write(\"\u6570\u636e\u89c4\u6a21: 10,000 \u7b14\u4ea4\u6613\uff0c\u5176\u4e2d 5% (500\u7b14) \u4e3a\u6b3a\u8bc8\u4ea4\u6613\u3002\\n\")\n f.write(\"\u5173\u952e\u5b57\u6bb5: \u4ea4\u6613ID, \u7528\u6237ID, \u5546\u54c1ID, \u4ea4\u6613\u91d1\u989d, \u65f6\u95f4\u6233, \u7528\u6237\u5386\u53f2\u884c\u4e3a, \u8bbe\u5907\u4fe1\u606f, IP\u5730\u5740, \u5730\u7406\u4f4d\u7f6e\u7b49\u3002\\n\")\n f.write(\"\u539f\u59cb\u6570\u636e\u5df2\u4fdd\u5b58\u4e3a CSV \u6587\u4ef6\u3002\\n\\n\")\n\n f.write(\"--- 3. \u65e0\u76d1\u7763\u5f02\u5e38\u68c0\u6d4b ---\\n\")\n f.write(\"\u76ee\u6807: \u5728\u6ca1\u6709\u6807\u7b7e\u7684\u60c5\u51b5\u4e0b\uff0c\u8bc6\u522b\u51fa\u4e0e\u5927\u591a\u6570\u6570\u636e\u663e\u8457\u4e0d\u540c\u7684\u5f02\u5e38\u70b9\u3002\\n\")\n f.write(\"\u65b9\u6cd5: Isolation Forest, Local Outlier Factor (LOF)\u3002\\n\")\n f.write(\"\u8bc4\u4f30: \u7531\u4e8e\u6ca1\u6709\u771f\u5b9e\u6807\u7b7e\uff0c\u8bc4\u4f30\u4e3b\u8981\u57fa\u4e8e\u6a21\u578b\u53d1\u73b0\u7684\u5f02\u5e38\u4e0e\u5df2\u77e5\u6b3a\u8bc8\u6837\u672c\u7684\u91cd\u5408\u5ea6\u3002\\n\\n\")\n \n f.write(\"\u5404\u6a21\u578b\u6027\u80fd\u5bf9\u6bd4:\\n\")\n unsup_comparison_df = pd.DataFrame({\n 'Model': list(unsup_results.keys()),\n 'Accuracy': [unsup_results[k]['accuracy'] for k in unsup_results.keys()],\n 'Precision': [unsup_results[k]['precision'] for k in unsup_results.keys()],\n 'Recall': [unsup_results[k]['recall'] for k in unsup_results.keys()],\n 'F1-Score': [unsup_results[k]['f1'] for k in unsup_results.keys()],\n 'AUC-ROC': [unsup_results[k]['auc'] for k in unsup_results.keys()]\n })\n f.write(unsup_comparison_df.round(4).to_string(index=False))\n f.write(\"\\n\\n\u6ce8\u610f\uff1a\u65e0\u76d1\u7763\u65b9\u6cd5\u7684\u7cbe\u786e\u7387\u548c\u53ec\u56de\u7387\u53ef\u80fd\u4e0d\u5982\u76d1\u7763\u65b9\u6cd5\uff0c\u4f46\u80fd\u53d1\u73b0\u672a\u77e5\u6a21\u5f0f\u3002\\n\\n\")\n\n f.write(\"--- 4. \u6709\u76d1\u7763\u6b3a\u8bc8\u68c0\u6d4b ---\\n\")\n f.write(\"\u76ee\u6807: \u5229\u7528\u5df2\u77e5\u7684\u6b3a\u8bc8\/\u6b63\u5e38\u6807\u7b7e\uff0c\u8bad\u7ec3\u5206\u7c7b\u6a21\u578b\u6765\u9884\u6d4b\u65b0\u4ea4\u6613\u662f\u5426\u4e3a\u6b3a\u8bc8\u3002\\n\")\n f.write(\"\u65b9\u6cd5: Random Forest (\u4f7f\u7528\u4e86\u7c7b\u522b\u6743\u91cd\u5904\u7406\u6837\u672c\u4e0d\u5e73\u8861)\u3002\\n\")\n f.write(\"\u8bc4\u4f30\u6307\u6807\u5305\u62ec: \u51c6\u786e\u7387 (Accuracy), \u7cbe\u786e\u7387 (Precision), \u53ec\u56de\u7387 (Recall), F1\u5206\u6570 (F1-Score), AUC-ROC\u3002\\n\\n\")\n \n f.write(f\"Random Forest \u6a21\u578b\u5728\u6d4b\u8bd5\u96c6\u4e0a\u7684\u8868\u73b0:\\n\")\n f.write(f\" - \u51c6\u786e\u7387 (Accuracy): {sup_metrics['accuracy']:.4f}\\n\")\n f.write(f\" - \u7cbe\u786e\u7387 (Precision): {sup_metrics['precision']:.4f}\\n\")\n f.write(f\" - \u53ec\u56de\u7387 (Recall): {sup_metrics['recall']:.4f}\\n\")\n f.write(f\" - F1\u5206\u6570 (F1-Score): {sup_metrics['f1']:.4f}\\n\")\n f.write(f\" - AUC-ROC: {sup_metrics['auc']:.4f}\\n\")\n f.write(\"\u6df7\u6dc6\u77e9\u9635\u8be6\u7ec6\u5206\u6790\u4e86\u6a21\u578b\u7684\u9884\u6d4b\u7ed3\u679c\uff0c\u56fe\u8868\u5df2\u751f\u6210\u3002\\n\")\n f.write(f\"\u6df7\u6dc6\u77e9\u9635\u56fe\u8868: {cm_path}\\n\\n\")\n\n f.write(\"--- 5. \u5173\u952e\u9a71\u52a8\u56e0\u7d20\u5206\u6790 ---\\n\")\n f.write(\"\u901a\u8fc7\u5206\u6790\u6700\u4f73\u6a21\u578b (Random Forest) \u7684\u7279\u5f81\u91cd\u8981\u6027\uff0c\u8bc6\u522b\u51fa\u533a\u5206\u6b3a\u8bc8\u548c\u6b63\u5e38\u4ea4\u6613\u7684\u6700\u5173\u952e\u56e0\u7d20\u3002\\n\")\n f.write(\"\u7279\u5f81\u91cd\u8981\u6027\u6392\u5e8f (Top 10):\\n\")\n f.write(feat_imp_df.head(10).to_string(index=False))\n f.write(\"\\n\u4ece\u4e0a\u8868\u53ef\u4ee5\u770b\u51fa\uff0c\u54ea\u4e9b\u4ea4\u6613\u7279\u5f81\u6700\u80fd\u6307\u793a\u6b3a\u8bc8\u884c\u4e3a\u3002\\n\")\n f.write(f\"\u7279\u5f81\u91cd\u8981\u6027\u56fe\u8868: {feat_imp_path}\\n\\n\")\n\n f.write(\"--- 6. \u98ce\u9669\u7ba1\u7406\u4e0e\u4e1a\u52a1\u5e94\u7528\u5efa\u8bae ---\\n\")\n f.write(\"1. \u5b9e\u65f6\u76d1\u63a7: \u5c06\u8bad\u7ec3\u597d\u7684\u6a21\u578b\u90e8\u7f72\u5230\u4ea4\u6613\u7cfb\u7edf\u4e2d\uff0c\u5bf9\u6bcf\u7b14\u5b9e\u65f6\u4ea4\u6613\u8fdb\u884c\u8bc4\u5206\u3002\\n\")\n f.write(\"2. \u98ce\u9669\u9608\u503c: \u6839\u636e\u4e1a\u52a1\u53ef\u63a5\u53d7\u7684\u8bef\u62a5\u7387\uff0c\u8bbe\u5b9a\u6b3a\u8bc8\u6982\u7387\u9608\u503c\u3002\u8d85\u8fc7\u9608\u503c\u7684\u4ea4\u6613\u8fdb\u5165\u4eba\u5de5\u5ba1\u6838\u6216\u88ab\u6682\u65f6\u963b\u6b62\u3002\\n\")\n f.write(\"3. \u591a\u5c42\u9632\u62a4:\\n\")\n f.write(\" - \u7ed3\u5408\u65e0\u76d1\u7763\u548c\u6709\u76d1\u7763\u6a21\u578b\u7684\u7ed3\u679c\uff0c\u63d0\u9ad8\u68c0\u6d4b\u7684\u9c81\u68d2\u6027\u3002\\n\")\n f.write(\" - \u5bf9\u4e8e\u9ad8\u98ce\u9669\u7279\u5f81\u7ec4\u5408\uff08\u5982'\u65b0\u8bbe\u5907'+'\u5927\u989d\u4ea4\u6613'+'\u975e\u5e38\u7528\u56fd\u5bb6'\uff09\uff0c\u5373\u4f7f\u6a21\u578b\u8bc4\u5206\u4e0d\u9ad8\u4e5f\u5e94\u52a0\u5f3a\u5173\u6ce8\u3002\\n\")\n f.write(\"4. \u89c4\u5219\u5f15\u64ce: \u5bf9\u4e8e\u4e00\u4e9b\u660e\u663e\u7684\u6b3a\u8bc8\u6a21\u5f0f\uff08\u5982IP\u5730\u5740\u9ed1\u540d\u5355\uff09\uff0c\u53ef\u4ee5\u4f7f\u7528\u89c4\u5219\u5f15\u64ce\u5feb\u901f\u62e6\u622a\u3002\\n\")\n f.write(\"5. \u6a21\u578b\u8fed\u4ee3: \u5b9a\u671f\u4f7f\u7528\u6700\u65b0\u7684\u5e26\u6807\u7b7e\u6570\u636e\u91cd\u65b0\u8bad\u7ec3\u6a21\u578b\uff0c\u4ee5\u9002\u5e94\u4e0d\u65ad\u53d8\u5316\u7684\u6b3a\u8bc8\u624b\u6bb5\u3002\\n\")\n f.write(\"6. \u7528\u6237\u6559\u80b2: \u63d0\u9192\u7528\u6237\u4fdd\u62a4\u8d26\u6237\u5b89\u5168\uff0c\u5982\u4e0d\u6cc4\u9732\u5bc6\u7801\u3001\u4e0d\u70b9\u51fb\u53ef\u7591\u94fe\u63a5\u7b49\u3002\\n\\n\")\n\n f.write(\"=\" * 50 + \"\\n\")\n f.write(\" \u62a5\u544a\u7ed3\u675f\\n\")\n f.write(\"=\" * 50 + \"\\n\")\n\n print(f\"\u6b3a\u8bc8\u68c0\u6d4b\u5206\u6790\u62a5\u544a\u5df2\u751f\u6210: {report_filename}\")\n\n# --- \u4e3b\u51fd\u6570 ---\n\ndef main():\n \"\"\"\u4e3b\u51fd\u6570\"\"\"\n # 1. \u751f\u6210\u6570\u636e\n df_fraud = generate_sample_fraud_data(NUM_TRANSACTIONS, FRAUD_RATIO)\n \n # 2. \u6570\u636e\u9884\u5904\u7406\n X, y, le_country, feature_cols = preprocess_data(df_fraud)\n \n # 3. \u65e0\u76d1\u7763\u5f02\u5e38\u68c0\u6d4b\n unsup_results = perform_unsupervised_detection(X, y)\n \n # 4. \u6709\u76d1\u7763\u6b3a\u8bc8\u68c0\u6d4b\n sup_results, cm_path = perform_supervised_detection(X, y)\n \n # 5. \u7279\u5f81\u91cd\u8981\u6027\u5206\u6790 (\u9488\u5bf9\u6709\u76d1\u7763\u6a21\u578b)\n best_model = sup_results['Random Forest']['model']\n feat_imp_df, feat_imp_path = analyze_feature_importance(best_model, feature_cols, 'Random Forest')\n \n # 6. \u751f\u6210\u62a5\u544a\n generate_fraud_detection_report(unsup_results, sup_results, cm_path, feat_imp_df, feat_imp_path)\n \n print(\"\\n\u7528\u6237\u98ce\u9669\u7ba1\u7406\u548c\u6b3a\u8bc8\u884c\u4e3a\u68c0\u6d4b\u5206\u6790\u6d41\u7a0b\u5b8c\u6210\u3002\")\n\nif __name__ == \"__main__\":\n main()<\/code><\/pre>\n","protected":false},"excerpt":{"rendered":"\u8be5\u811a\u672c\u5c06\u5305\u542b\u4ee5\u4e0b\u529f\u80fd\uff1a\u4f7f\u7528\u968f\u673a\u68ee\u6797\uff08Random Forest\uff09\u8fdb\u884c\u6709\u76d1\u7763\u7684\u6b3a\u8bc8\u68c0\u6d4b<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":[],"categories":[1],"tags":[28],"views":351,"_links":{"self":[{"href":"http:\/\/viplao.com\/index.php\/wp-json\/wp\/v2\/posts\/3897"}],"collection":[{"href":"http:\/\/viplao.com\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"http:\/\/viplao.com\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"http:\/\/viplao.com\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"http:\/\/viplao.com\/index.php\/wp-json\/wp\/v2\/comments?post=3897"}],"version-history":[{"count":2,"href":"http:\/\/viplao.com\/index.php\/wp-json\/wp\/v2\/posts\/3897\/revisions"}],"predecessor-version":[{"id":3913,"href":"http:\/\/viplao.com\/index.php\/wp-json\/wp\/v2\/posts\/3897\/revisions\/3913"}],"wp:attachment":[{"href":"http:\/\/viplao.com\/index.php\/wp-json\/wp\/v2\/media?parent=3897"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"http:\/\/viplao.com\/index.php\/wp-json\/wp\/v2\/categories?post=3897"},{"taxonomy":"post_tag","embeddable":true,"href":"http:\/\/viplao.com\/index.php\/wp-json\/wp\/v2\/tags?post=3897"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}