银行风控模型
一、神经网络
1 import pandas as pd 2 import numpy as np 3 from sklearn.model_selection import train_test_split 4 datafile = 'bankloan2.xls' 5 data = pd.read_excel(datafile) 6 x = data.iloc[:,:8] 7 y = data.iloc[:,8] 8 x_train, x_test, y_train, y_test = train_test_split(x, y, test_size=0.2, random_state=100) 9 from tensorflow.keras.models import Sequential 10 from tensorflow.keras.layers import Dense,Dropout 11 from tensorflow.keras.metrics import BinaryAccuracy 12 import time 13 start_time = time.time() 14 model = Sequential() 15 model.add(Dense(input_dim=8,units=800,activation='relu')) 16 model.add(Dropout(0.5)) 17 model.add(Dense(input_dim=800,units=400,activation='relu')) 18 model.add(Dropout(0.5)) 19 # model.add(Dense(input_dim=800,units=400,activation='relu')) 20 # model.add(Dropout(0.5)) 21 # model.add(Dense(input_dim=400,units=200,activation='softsign')) 22 # model.add(Dropout(0.5)) 23 model.add(Dense(input_dim=400,units=1,activation='sigmoid')) 24 25 model.compile(loss='binary_crossentropy', optimizer='adam',metrics=[BinaryAccuracy()]) 26 model.fit(x_train,y_train,epochs=500,batch_size=128) 27 loss,binary_accuracy = model.evaluate(x,y,batch_size=128) 28 end_time = time.time() 29 run_time = end_time-start_time 30 print('模型运行时间：{}'.format(run_time)) 31 print('模型损失值：{}'.format(loss)) 32 print('模型精度：{}'.format(binary_accuracy)) 33 34 yp = model.predict(x).reshape(len(y)) 35 yp = np.around(yp,0).astype(int) #转换为整型 36 from cm_plot import * # 导入自行编写的混淆矩阵可视化函数 37 38 cm_plot(y,yp).show() # 显示混淆矩阵可视化结果
混淆矩阵可视化函数cm_plot.py
1 def cm_plot(y, yp): 2 3 from sklearn.metrics import confusion_matrix #导入混淆矩阵函数 4 5 cm = confusion_matrix(y, yp) #混淆矩阵 6 7 import matplotlib.pyplot as plt #导入作图库 8 plt.matshow(cm, cmap=plt.cm.Greens) #画混淆矩阵图，配色风格使用cm.Greens，更多风格请参考官网。 9 plt.colorbar() #颜色标签 10 11 for x in range(len(cm)): #数据标签 12 for y in range(len(cm)): 13 plt.annotate(cm[x,y], xy=(x, y), horizontalalignment='center', verticalalignment='center') 14 15 plt.ylabel('True label') #坐标轴标签 16 plt.xlabel('Predicted label') #坐标轴标签 17 return plt
训练结果
二、用支持向量机、决策树、随机森林方法训练
代码
1 import pandas as pd 2 import time 3 import numpy as np 4 import seaborn as sns 5 import matplotlib.pyplot as plt 6 from sklearn.model_selection import train_test_split 7 from sklearn.tree import DecisionTreeClassifier as DTC 8 from sklearn.ensemble import RandomForestClassifier as RFC 9 from sklearn import svm 10 from sklearn import tree 11 from sklearn.metrics import confusion_matrix 12 from sklearn.metrics import accuracy_score 13 from sklearn.metrics import roc_curve, auc 14 from sklearn.neighbors import KNeighborsClassifier as KNN 15 #导入plot_roc_curve,roc_curve和roc_auc_score模块 16 from sklearn.metrics import plot_roc_curve,roc_curve,auc,roc_auc_score 17 filePath = 'bankloan2.xls' 18 data = pd.read_excel(filePath) 19 x = data.iloc[:,:8] 20 y = data.iloc[:,8] 21 x_train, x_test, y_train, y_test = train_test_split(x, y, test_size=0.2, random_state=100) 22 23 #模型 24 svm_clf = svm.SVC() 25 dtc_clf = DTC(criterion='entropy') 26 rfc_clf = RFC(n_estimators=10) 27 knn_clf = KNN() 28 29 #训练 30 knn_clf.fit(x_train,y_train) 31 rfc_clf.fit(x_train,y_train) 32 dtc_clf.fit(x_train,y_train) 33 svm_clf.fit(x_train, y_train) 34 35 36 #ROC曲线比较 37 fig,ax = plt.subplots(figsize=(12,10)) 38 rfc_roc = plot_roc_curve(estimator=rfc_clf, X=x, 39 y=y, ax=ax, linewidth=1) 40 svm_roc = plot_roc_curve(estimator=svm_clf, X=x, 41 y=y, ax=ax, linewidth=1) 42 dtc_roc = plot_roc_curve(estimator=dtc_clf, X=x, 43 y=y, ax=ax, linewidth=1) 44 knn_roc = plot_roc_curve(estimator=knn_clf, X=x, 45 y=y, ax=ax, linewidth=1) 46 ax.legend(fontsize=12) 47 plt.show() 48 49 #模型评价 50 rfc_yp = rfc_clf.predict(x) 51 rfc_score = accuracy_score(y, rfc_yp) 52 svm_yp = svm_clf.predict(x) 53 svm_score = accuracy_score(y, svm_yp) 54 dtc_yp = dtc_clf.predict(x) 55 dtc_score = accuracy_score(y, dtc_yp) 56 knn_yp = knn_clf.predict(x) 57 knn_score = accuracy_score(y, knn_yp) 58 score = {"随机森林得分":rfc_score,"支持向量机得分":svm_score,"决策树得分":dtc_score,"K邻近得分":knn_score} 59 score = sorted(score.items(),key = lambda score:score[0],reverse=True) 60 print(pd.DataFrame(score)) 61 62 #中文标签、负号正常显示 63 plt.rcParams['font.sans-serif'] = ['SimHei'] 64 plt.rcParams['axes.unicode_minus'] = False 65 66 #绘制混淆矩阵 67 figure = plt.subplots(figsize=(12,10)) 68 plt.subplot(2,2,1) 69 plt.title('随机森林') 70 rfc_cm = confusion_matrix(y, rfc_yp) 71 heatmap = sns.heatmap(rfc_cm, annot=True, fmt='d') 72 heatmap.yaxis.set_ticklabels(heatmap.yaxis.get_ticklabels(), rotation=0, ha='right') 73 heatmap.xaxis.set_ticklabels(heatmap.xaxis.get_ticklabels(), rotation=45, ha='right') 74 plt.ylabel("true label") 75 plt.xlabel("predict label") 76 77 plt.subplot(2,2,2) 78 plt.title('支持向量机') 79 svm_cm = confusion_matrix(y, svm_yp) 80 heatmap = sns.heatmap(svm_cm, annot=True, fmt='d') 81 heatmap.yaxis.set_ticklabels(heatmap.yaxis.get_ticklabels(), rotation=0, ha='right') 82 heatmap.xaxis.set_ticklabels(heatmap.xaxis.get_ticklabels(), rotation=45, ha='right') 83 plt.ylabel("true label") 84 plt.xlabel("predict label") 85 86 plt.subplot(2,2,3) 87 plt.title('决策树') 88 dtc_cm = confusion_matrix(y, dtc_yp) 89 heatmap = sns.heatmap(dtc_cm, annot=True, fmt='d') 90 heatmap.yaxis.set_ticklabels(heatmap.yaxis.get_ticklabels(), rotation=0, ha='right') 91 heatmap.xaxis.set_ticklabels(heatmap.xaxis.get_ticklabels(), rotation=45, ha='right') 92 plt.ylabel("true label") 93 plt.xlabel("predict label") 94 95 plt.subplot(2,2,4) 96 plt.title('K邻近') 97 knn_cm = confusion_matrix(y, knn_yp) 98 heatmap = sns.heatmap(knn_cm, annot=True, fmt='d') 99 heatmap.yaxis.set_ticklabels(heatmap.yaxis.get_ticklabels(), rotation=0, ha='right') 100 heatmap.xaxis.set_ticklabels(heatmap.xaxis.get_ticklabels(), rotation=45, ha='right') 101 plt.ylabel("true label") 102 plt.xlabel("predict label") 103 plt.show() 104 105 106 #画出决策树 107 import pandas as pd 108 from sklearn.tree import export_graphviz 109 x = pd.DataFrame(x) 110 111 with open(r"banklodan_tree1.dot", 'w') as f: 112 export_graphviz(dtc_clf, feature_names = x.columns, out_file = f) 113 f.close() 114 115 from IPython.display import Image 116 from sklearn import tree 117 import pydotplus 118 dot_data = tree.export_graphviz(dtc_clf, out_file=None, #regr_1 是对应分类器 119 feature_names=x.columns, #对应特征的名字 120 class_names= ['不违约','违约'], #对应类别的名字 121 filled=True, rounded=True, 122 special_characters=True)
结果
（1）混淆矩阵
（2）ROC曲线
（3）