1.9 金融风控实战：产品购买预测、客户流失预警、反欺诈模型构建
引言
金融风控是数据挖掘在金融领域最重要的应用之一。本文将实战演示三个核心风控场景：产品购买预测、客户流失预警、反欺诈模型构建。通过完整的代码实现，带你从数据到模型，掌握金融风控的实战技能。
一、场景1：产品购买预测实战
1.1 业务背景与数据准备
importpandasaspdimportnumpyasnpfromsklearn.model_selectionimporttrain_test_splitfromsklearn.ensembleimportGradientBoostingClassifierfromsklearn.metricsimportroc_auc_score,classification_report,confusion_matrixclassProductPurchasePrediction:"""产品购买预测模型 - 完整实战"""def__init__(self):self.model=Noneself.feature_importance=Nonedefgenerate_data(self,n_samples=10000):"""生成模拟数据"""print("="*60)print("场景1: 产品购买预测")print("="*60)np.random.seed(42)data={'customer_id':range(1,n_samples+1),'age':np.random.randint(18,70,n_samples),'income':np.random.uniform(3000,50000,n_samples),'credit_score':np.random.randint(300,850,n_samples),'existing_products':np.random.randint(0,5,n_samples),'transaction_count':np.random.poisson(10,n_samples),'avg_transaction_amount':np.random.uniform(100,5000,n_samples),'days_since_last_transaction':np.random.randint(0,90,n_samples),'customer_lifetime_value':np.random.uniform(1000,100000,n_samples),'product_interest_score':np.random.uniform(0,1,n_samples)}df=pd.DataFrame(data)# 生成购买标签（基于业务规则）purchase_prob=((df['income']>20000)*0.15+(df['credit_score']>700)*0.12+(df['product_interest_score']>0.7)*0.20+(df['customer_lifetime_value']>50000)*0.10+(df['transaction_count']>15)*0.08+np.random.uniform(0,0.35,n_samples))df['will_purchase']=(purchase_prob>0.5).astype(int)print(f"\n数据生成完成:")print(f" 样本数:{len(df):,}")print(f" 特征数:{len(df.columns)-1}")print(f" 购买率:{df['will_purchase'].mean():.2%}")returndfdeffeature_engineering(self,df):"""特征工程"""print("\n特征工程")# 创建新特征df['income_to_age_ratio']=df['income']/(df['age']+1)df['high_value_customer']=(df['customer_lifetime_value']>50000).astype(int)df['active_customer']=(df['transaction_count']>15).astype(int)df['recent_activity']=(df['days_since_last_transaction']<30).astype(int)print(f" 创建了 4 个新特征")print(f" 总特征数:{len(df.columns)-1}")returndfdeftrain_model(self,df):"""训练模型"""print("\n模型训练")# 准备特征feature_cols=[colforcolindf.columnsifcolnotin['customer_id','will_purchase']]X=df[feature_cols]y=df['will_purchase']# 划分数据X_train,X_test,y_train,y_test=train_test_split(X,y,test_size=0.2,random_state=42,stratify=y)print(f" 训练集:{len(X_train):,}样本")print(f" 测试集:{len(X_test):,}样本")# 训练模型self.model=GradientBoostingClassifier(n_estimators=100,learning_rate=0.1,max_depth=5,min_samples_split=100,random_state=42)self.model.fit(X_train,y_train)# 预测y_pred=self.model.predict(X_test)y_pred_proba=self.model.predict_proba(X_test)[:,1]# 评估auc=roc_auc_score(y_test,y_pred_proba)print(f"\n模型性能:")print(f" AUC:{auc:.4f}")print(f"\n分类报告:")print(classification_report(y_test,y_pred))# 特征重要性self.feature_importance=pd.DataFrame({'feature':feature_cols,'importance':self.model.feature_importances_}).sort_values('importance',ascending=False)print(f"\nTop 5 重要特征:")print(self.feature_importance.head().to_string(index=False))returnself.model,aucdefbusiness_application
