模型的评价

from sklearn.metrics import accuracy_score, recall_score, f1_score, roc_auc_score, roc_curve, precision_score
from matplotlib import pyplot as plt
# 定义评估函数
def model_metrics(clf, X_train, X_test, y_train, y_test):
    # 预测
    y_train_pred = clf.predict(X_train)
    y_test_pred = clf.predict(X_test)
    y_train_pred_proba = clf.predict_proba(X_train)[:, 1]
    y_test_pred_proba = clf.predict_proba(X_test)[:, 1]
 
    # 评估
    # 准确性
    print('准确性：')
    print('Train：{:.4f}'.format(accuracy_score(y_train, y_train_pred)))
    print('Test：{:.4f}'.format(accuracy_score(y_test, y_test_pred)))
 
    #精确性
    print('精确性：')
    print("Train:{:.4f}".format(precision_score(y_train, y_train_pred)))
    print("Test: {:.4f}".format(precision_score(y_train, y_train_pred)))
    # 召回率
    print('召回率：')
    print('Train：{:.4f}'.format(recall_score(y_train, y_train_pred)))
    print('Test：{:.4f}'.format(recall_score(y_test, y_test_pred)))
    
    # f1_score
    print('f1_score：')
    print('Train：{:.4f}'.format(f1_score(y_train, y_train_pred)))
    print('Test：{:.4f}'.format(f1_score(y_test, y_test_pred)))
    
    # roc_auc
    print('roc_auc：')
    print('Train：{:.4f}'.format(roc_auc_score(y_train, y_train_pred_proba)))
    print('Test：{:.4f}'.format(roc_auc_score(y_test, y_test_pred_proba)))    
    
    # 描绘 ROC 曲线
    fpr_tr, tpr_tr, _ = roc_curve(y_train, y_train_pred_proba)
    fpr_te, tpr_te, _ = roc_curve(y_test, y_test_pred_proba)
 
    # KS
    print('KS：')
    print('Train：{:.4f}'.format(max(abs((fpr_tr - tpr_tr)))))
    print('Test：{:.4f}'.format(max(abs((fpr_te - tpr_te)))))
 
    # 绘图
    plt.plot(fpr_tr, tpr_tr, 'r-',
             label="Train:AUC: {:.3f} KS:{:.3f}".format(roc_auc_score(y_train, y_train_pred_proba), 
                                                        max(abs((fpr_tr - tpr_tr)))))
    plt.plot(fpr_te, tpr_te, 'g-',
             label="Test:AUC: {:.3f} KS:{:.3f}".format(roc_auc_score(y_test, y_test_pred_proba),
                                                     max(abs((fpr_tr - tpr_tr)))))
    plt.plot([0, 1], [0, 1], 'd--')
    plt.legend(loc='best')
    plt.title("ROC curse")
    plt.show()

- 每个模型的准确率，精确率，召回率，F1-score，AUC值，ROC曲线

模型	准确率	精确率	召回率	F1-score	AUC值	ROC曲线
逻辑回归	train 0.7493 test 0.7477	train 0.0000 test 0.0000	train 0.0000 test 0.0000	train 0.0000 test 0.0000	train 0.5954 test 0.5838
决策树	train 1.0000 test 0.6847	train 1.0000 test 1.000	train 1.0000 test 0.4.95	train 1.0000 test 0.3952	train1.0000 test 0.5933
SVM	train 1.0000 test 0.7477	train 0.000 test 0.000	train 0.000 test 0.000	train 0.000 test 0.000	0.5954 test 0.5838
随机森林	train 0.9862 test 0.7694	train 0.9937 test 0.9937	train 0.9508 test 0.2841	train 0.9718 test 0.3827	0.9988 test 0.7374
XGBoost	train 0.8602 test 0.7933	train 0.8625 test 0.8625	train 0.5264 test 0.3928	train 0.6538 test 0.4887	0.9222 test 0.7788

• reference
  https://blog.csdn.net/l75326747/article/details/84233247