【Python】基于机器学习的财务数据分析——识别财务造假

【Python】基于机器学习的财务数据分析——识别财务造假

前言：

本文数据使用了2021泰迪杯官方给出的数据。

其中第一章的代码给出了如何由比赛数据生成案例分析所使用的数据

而第二章则重点介绍了 如何通过上一章的数据进行财务数据分析

---

第一章 生成财务分析数据

1.1 Load 比赛的官方数据

# load lab
import pandas as pd
import numpy as np


# load data
df = pd.read_csv('附件2(样例数据).csv')

df.info()
df

	TICKER_SYMBOL	ACT_PUBTIME	PUBLISH_DATE	END_DATE_REP	END_DATE	REPORT_TYPE	FISCAL_PERIOD	MERGED_FLAG	ACCOUTING_STANDARDS	CURRENCY_CD	...	CA_TURNOVER	OPER_CYCLE	INVEN_TURNOVER	FA_TURNOVER	TFA_TURNOVER	DAYS_AP	DAYS_INVEN	TA_TURNOVER	AR_TURNOVER	FLAG
0	4019	5	5	4	3	A	12	1	CHAS_2007	CNY	...	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	0
1	4213	5	5	5	3	A	12	1	CHAS_2007	CNY	...	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	0
2	8166	5	5	3	3	A	12	1	CHAS_2007	CNY	...	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	0
3	9063	5	5	4	3	A	12	1	CHAS_2007	CNY	...	5.3401	3.2947	NaN	4009.2584	4402.6179	1.2731	NaN	4.6236	109.2667	0
4	10083	5	5	4	3	A	12	1	CHAS_2007	CNY	...	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	0
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
3558	4992858	5	5	4	3	A	12	1	CHAS_2007	CNY	...	0.0601	6785.7245	0.0535	32.8540	94.7623	76.3016	6732.3615	0.0563	6.7463	0
3559	4993201	4	4	3	3	A	12	1	CHAS_2007	CNY	...	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	0
3560	4993297	6	6	5	3	A	12	1	CHAS_2007	CNY	...	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	0
3561	4998808	5	5	4	3	A	12	1	CHAS_2007	CNY	...	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	NaN	0
3562	4999709	5	5	4	3	A	12	1	CHAS_2007	CNY	...	1.2648	96.8794	4.7511	0.2166	0.2019	169.4288	75.7725	0.1492	17.0560	0

3563 rows × 363 columns

1.2 数据处理

df1 = df.copy()
df2 = df1.dropna(axis=1)
df2.dropna(axis=0)

	TICKER_SYMBOL	ACT_PUBTIME	PUBLISH_DATE	END_DATE_REP	END_DATE	REPORT_TYPE	FISCAL_PERIOD	MERGED_FLAG	ACCOUTING_STANDARDS	CURRENCY_CD	...	REVENUE	T_REVENUE	T_PROFIT	OPERATE_PROFIT	COMPR_INC_ATTR_P	T_COMPR_INCOME	N_INCOME_ATTR_P	N_INCOME	T_COGS	FLAG
0	4019	5	5	4	3	A	12	1	CHAS_2007	CNY	...	6.144704e+08	6.144704e+08	-1.081696e+08	-1.229547e+08	-9.021130e+07	-9.021130e+07	-9.021130e+07	-9.021130e+07	7.362240e+08	0
1	4213	5	5	5	3	A	12	1	CHAS_2007	CNY	...	1.103340e+09	1.103340e+09	1.440264e+08	1.417321e+08	1.173408e+08	1.173408e+08	1.173408e+08	1.173408e+08	9.609019e+08	0
2	8166	5	5	3	3	A	12	1	CHAS_2007	CNY	...	6.768557e+09	6.768557e+09	5.501612e+08	4.160823e+08	4.427106e+08	4.427106e+08	4.406464e+08	4.406464e+08	6.374538e+09	0
3	9063	5	5	4	3	A	12	1	CHAS_2007	CNY	...	3.649249e+10	3.649249e+10	4.442545e+08	4.178863e+08	3.603373e+08	3.590776e+08	3.582523e+08	3.569925e+08	3.610338e+10	0
4	10083	5	5	4	3	A	12	1	CHAS_2007	CNY	...	4.090022e+08	4.090022e+08	5.522668e+07	5.422161e+07	4.059508e+07	4.059508e+07	4.059508e+07	4.059508e+07	3.547806e+08	0
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
3558	4992858	5	5	4	3	A	12	1	CHAS_2007	CNY	...	1.916369e+08	1.916369e+08	-5.854943e+07	-6.549837e+07	-5.698415e+07	-5.888314e+07	-5.698415e+07	-5.888314e+07	2.563552e+08	0
3559	4993201	4	4	3	3	A	12	1	CHAS_2007	CNY	...	7.584324e+08	7.584324e+08	4.562972e+07	4.016267e+07	3.719758e+07	3.704166e+07	3.719758e+07	3.704166e+07	7.174872e+08	0
3560	4993297	6	6	5	3	A	12	1	CHAS_2007	CNY	...	8.707040e+08	8.707040e+08	1.657144e+08	1.204156e+08	1.428394e+08	1.623189e+08	1.422310e+08	1.615584e+08	7.586088e+08	0
3561	4998808	5	5	4	3	A	12	1	CHAS_2007	CNY	...	1.345086e+09	1.345086e+09	8.585497e+07	7.220415e+07	7.851344e+07	7.895948e+07	7.500529e+07	7.545319e+07	1.275262e+09	0
3562	4999709	5	5	4	3	A	12	1	CHAS_2007	CNY	...	1.260865e+10	1.260865e+10	3.371832e+09	3.297172e+09	2.258473e+09	2.677414e+09	2.460094e+09	2.870661e+09	1.165588e+10	0

3563 rows × 34 columns

df2.to_excel('winner.xlsx',index = False)

df3 = df2.iloc[:,11:]
df3

	FIXED_ASSETS	T_ASSETS	T_LIAB	T_EQUITY_ATTR_P	T_SH_EQUITY	T_LIAB_EQUITY	N_CF_OPERATE_A	C_PAID_FOR_OTH_OP_A	C_INF_FR_OPERATE_A	N_CHANGE_IN_CASH	...	REVENUE	T_REVENUE	T_PROFIT	OPERATE_PROFIT	COMPR_INC_ATTR_P	T_COMPR_INCOME	N_INCOME_ATTR_P	N_INCOME	T_COGS	FLAG
0	1.384757e+09	3.360603e+09	9.086384e+08	2.451965e+09	2.451965e+09	3.360603e+09	-1.531725e+08	8.092480e+07	5.484389e+08	-5.863646e+07	...	6.144704e+08	6.144704e+08	-1.081696e+08	-1.229547e+08	-9.021130e+07	-9.021130e+07	-9.021130e+07	-9.021130e+07	7.362240e+08	0
1	3.550098e+08	1.801029e+09	4.622409e+08	1.338788e+09	1.338788e+09	1.801029e+09	1.530939e+08	3.725100e+07	1.059183e+09	-2.576530e+07	...	1.103340e+09	1.103340e+09	1.440264e+08	1.417321e+08	1.173408e+08	1.173408e+08	1.173408e+08	1.173408e+08	9.609019e+08	0
2	1.014511e+09	9.033789e+09	4.241620e+09	4.792168e+09	4.792168e+09	9.033789e+09	8.985383e+08	1.011182e+09	7.778336e+09	-7.102159e+08	...	6.768557e+09	6.768557e+09	5.501612e+08	4.160823e+08	4.427106e+08	4.427106e+08	4.406464e+08	4.406464e+08	6.374538e+09	0
3	2.592562e+09	1.445027e+10	9.246911e+09	5.049558e+09	5.203356e+09	1.445027e+10	4.287553e+08	3.640225e+09	4.261880e+10	2.915535e+08	...	3.649249e+10	3.649249e+10	4.442545e+08	4.178863e+08	3.603373e+08	3.590776e+08	3.582523e+08	3.569925e+08	3.610338e+10	0
4	4.234430e+07	5.389908e+08	2.419290e+08	2.970618e+08	2.970618e+08	5.389908e+08	3.351069e+07	2.934517e+07	4.872505e+08	1.073503e+07	...	4.090022e+08	4.090022e+08	5.522668e+07	5.422161e+07	4.059508e+07	4.059508e+07	4.059508e+07	4.059508e+07	3.547806e+08	0
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
3558	1.542683e+08	7.881017e+08	3.139182e+08	4.168727e+08	4.741834e+08	7.881017e+08	-5.238869e+06	1.838851e+07	1.844894e+08	-9.722585e+07	...	1.916369e+08	1.916369e+08	-5.854943e+07	-6.549837e+07	-5.698415e+07	-5.888314e+07	-5.698415e+07	-5.888314e+07	2.563552e+08	0
3559	2.945962e+08	8.894905e+08	2.485487e+08	6.370977e+08	6.409417e+08	8.894905e+08	8.333100e+07	7.960547e+07	8.047397e+08	-8.111294e+07	...	7.584324e+08	7.584324e+08	4.562972e+07	4.016267e+07	3.719758e+07	3.704166e+07	3.719758e+07	3.704166e+07	7.174872e+08	0
3560	3.435635e+09	5.628529e+09	2.599046e+09	2.774693e+09	3.029483e+09	5.628529e+09	4.344513e+08	2.430308e+07	1.084828e+09	8.119653e+07	...	8.707040e+08	8.707040e+08	1.657144e+08	1.204156e+08	1.428394e+08	1.623189e+08	1.422310e+08	1.615584e+08	7.586088e+08	0
3561	1.963836e+09	4.936854e+09	3.018880e+09	1.878145e+09	1.917974e+09	4.936854e+09	7.512584e+07	2.379392e+08	1.873552e+09	1.619822e+08	...	1.345086e+09	1.345086e+09	8.585497e+07	7.220415e+07	7.851344e+07	7.895948e+07	7.500529e+07	7.545319e+07	1.275262e+09	0
3562	4.760795e+09	3.820173e+10	1.753208e+10	1.818157e+10	2.066965e+10	3.820173e+10	1.621028e+09	3.193049e+09	1.472222e+10	3.384381e+08	...	1.260865e+10	1.260865e+10	3.371832e+09	3.297172e+09	2.258473e+09	2.677414e+09	2.460094e+09	2.870661e+09	1.165588e+10	0

3563 rows × 23 columns

df3.to_excel('raw_data.xlsx',index = False)

第二章 基于机器学习的财务数据分析——识别财务造假

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt

df = pd.read_excel('raw_data.xlsx')
df

	FIXED_ASSETS	T_ASSETS	T_LIAB	T_EQUITY_ATTR_P	T_SH_EQUITY	T_LIAB_EQUITY	N_CF_OPERATE_A	C_PAID_FOR_OTH_OP_A	C_INF_FR_OPERATE_A	N_CHANGE_IN_CASH	...	REVENUE	T_REVENUE	T_PROFIT	OPERATE_PROFIT	COMPR_INC_ATTR_P	T_COMPR_INCOME	N_INCOME_ATTR_P	N_INCOME	T_COGS	FLAG
0	1.384757e+09	3.360603e+09	9.086384e+08	2.451965e+09	2.451965e+09	3.360603e+09	-1.531725e+08	8.092480e+07	5.484389e+08	-5.863646e+07	...	6.144704e+08	6.144704e+08	-1.081696e+08	-1.229547e+08	-9.021130e+07	-9.021130e+07	-9.021130e+07	-9.021130e+07	7.362240e+08	0
1	3.550098e+08	1.801029e+09	4.622409e+08	1.338788e+09	1.338788e+09	1.801029e+09	1.530939e+08	3.725100e+07	1.059183e+09	-2.576530e+07	...	1.103340e+09	1.103340e+09	1.440264e+08	1.417321e+08	1.173408e+08	1.173408e+08	1.173408e+08	1.173408e+08	9.609019e+08	0
2	1.014511e+09	9.033789e+09	4.241620e+09	4.792168e+09	4.792168e+09	9.033789e+09	8.985383e+08	1.011182e+09	7.778336e+09	-7.102159e+08	...	6.768557e+09	6.768557e+09	5.501612e+08	4.160823e+08	4.427106e+08	4.427106e+08	4.406464e+08	4.406464e+08	6.374538e+09	0
3	2.592562e+09	1.445027e+10	9.246911e+09	5.049558e+09	5.203356e+09	1.445027e+10	4.287553e+08	3.640225e+09	4.261880e+10	2.915535e+08	...	3.649249e+10	3.649249e+10	4.442545e+08	4.178863e+08	3.603373e+08	3.590776e+08	3.582523e+08	3.569925e+08	3.610338e+10	0
4	4.234430e+07	5.389908e+08	2.419290e+08	2.970618e+08	2.970618e+08	5.389908e+08	3.351069e+07	2.934517e+07	4.872505e+08	1.073503e+07	...	4.090022e+08	4.090022e+08	5.522668e+07	5.422161e+07	4.059508e+07	4.059508e+07	4.059508e+07	4.059508e+07	3.547806e+08	0
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
3558	1.542683e+08	7.881017e+08	3.139182e+08	4.168727e+08	4.741834e+08	7.881017e+08	-5.238869e+06	1.838851e+07	1.844894e+08	-9.722585e+07	...	1.916369e+08	1.916369e+08	-5.854943e+07	-6.549837e+07	-5.698415e+07	-5.888314e+07	-5.698415e+07	-5.888314e+07	2.563552e+08	0
3559	2.945962e+08	8.894905e+08	2.485487e+08	6.370977e+08	6.409417e+08	8.894905e+08	8.333100e+07	7.960547e+07	8.047397e+08	-8.111294e+07	...	7.584324e+08	7.584324e+08	4.562972e+07	4.016267e+07	3.719758e+07	3.704166e+07	3.719758e+07	3.704166e+07	7.174872e+08	0
3560	3.435635e+09	5.628529e+09	2.599046e+09	2.774693e+09	3.029483e+09	5.628529e+09	4.344513e+08	2.430308e+07	1.084828e+09	8.119653e+07	...	8.707040e+08	8.707040e+08	1.657144e+08	1.204156e+08	1.428394e+08	1.623189e+08	1.422310e+08	1.615584e+08	7.586088e+08	0
3561	1.963836e+09	4.936854e+09	3.018880e+09	1.878145e+09	1.917974e+09	4.936854e+09	7.512584e+07	2.379392e+08	1.873552e+09	1.619822e+08	...	1.345086e+09	1.345086e+09	8.585497e+07	7.220415e+07	7.851344e+07	7.895948e+07	7.500529e+07	7.545319e+07	1.275262e+09	0
3562	4.760795e+09	3.820173e+10	1.753208e+10	1.818157e+10	2.066965e+10	3.820173e+10	1.621028e+09	3.193049e+09	1.472222e+10	3.384381e+08	...	1.260865e+10	1.260865e+10	3.371832e+09	3.297172e+09	2.258473e+09	2.677414e+09	2.460094e+09	2.870661e+09	1.165588e+10	0

3563 rows × 23 columns

X, y = df.iloc[:,:-1],df.iloc[:,-1]
y

0       0
1       0
2       0
3       0
4       0
       ..
3558    0
3559    0
3560    0
3561    0
3562    0
Name: FLAG, Length: 3563, dtype: int64

from sklearn.model_selection import train_test_split
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)
# 对数据进行标准化处理， 主要是X_train
from sklearn.preprocessing import StandardScaler
ss = StandardScaler()
ss = ss.fit(X_train)
X_train_std = ss.fit_transform(X_train)
X_test_std = ss.fit_transform(X_test)

#---------------  Modllong

#---------------  Modllong

# SVM Classifier  
def svm_classifier(train_x, train_y):  
    from sklearn.svm import SVC  
    model = SVC(kernel='rbf', probability=True)  
    model.fit(train_x, train_y)  
    return model 


# KNN Classifier  
def knn_classifier(train_x, train_y):  
    from sklearn.neighbors import KNeighborsClassifier  
    model = KNeighborsClassifier()  
    model.fit(train_x, train_y)  
    return model  
    
# Logistic Regression Classifier  
def logistic_regression_classifier(train_x, train_y):  
    from sklearn.linear_model import LogisticRegression  
    model = LogisticRegression(penalty='l2')  
    model.fit(train_x, train_y)  
    return model 

# DT
def dt(train_x, train_y):  
    from sklearn import tree
    model = tree.DecisionTreeClassifier()
    model.fit(train_x, train_y)  
    return model

# nn
def nn(train_x, train_y):  
    from sklearn.neural_network import MLPClassifier
    model = MLPClassifier(solver='lbfgs', alpha=1e-5,hidden_layer_sizes=(5, 2), random_state=1)
    model.fit(train_x, train_y)  
    return model


train_x = X_train_std
train_y = y_train
model_svc = svm_classifier(train_x, train_y)
model_knn = knn_classifier(train_x, train_y)
model_logistic =  logistic_regression_classifier(train_x, train_y)
model_dt = dt(train_x, train_y)
model_nn = nn(train_x, train_y)
# ----------
y_svc = model_svc.predict(X_test_std)
y_knn = model_knn.predict(X_test_std)
y_logistic = model_logistic.predict(X_test_std)
y_dt = model_dt.predict(X_test_std)
y_nn = model_nn.predict(X_test_std)
# 结果分析

from sklearn.metrics import f1_score
from sklearn.metrics import accuracy_score
from sklearn.metrics import  precision_score
from sklearn.metrics import recall_score

# print('分类准确率为：',accuracy_score(y_test,y_svc),accuracy_score(y_test,y_knn),accuracy_score(y_test,y_logistic),accuracy_score(y_test,y_rf))
# print('宏平均准确率：',precision_score(y_test,y_svc,average='macro'),precision_score(y_test,y_knn,average='macro'),precision_score(y_test,y_logistic,average='macro'),precision_score(y_test,y_rf,average='macro'))
# print('微平均准确率：',precision_score(y_test,y_svc,average='micro'),precision_score(y_test,y_knn,average='micro'),precision_score(y_test,y_logistic,average='micro'),precision_score(y_test,y_rf,average='micro'))
# print('宏平均召回率为：',recall_score(y_test,y_svc,average='macro'),recall_score(y_test,y_knn,average='macro'),recall_score(y_test,y_logistic,average='macro'),recall_score(y_test,y_rf,average='macro'))
# print('微平均召回率为：',recall_score(y_test,y_svc,average='micro'),recall_score(y_test,y_knn,average='micro'),recall_score(y_test,y_logistic,average='micro'),recall_score(y_test,y_rf,average='micro'))
# print('宏平均f1值为：',f1_score(y_test,y_svc,average='macro'),f1_score(y_test,y_knn,average='macro'),f1_score(y_test,y_logistic,average='macro'),f1_score(y_test,y_rf,average='macro'))
# print('微平均f1值为：',f1_score(y_test,y_svc,average='micro'),f1_score(y_test,y_knn,average='micro'),f1_score(y_test,y_logistic,average='micro'),f1_score(y_test,y_rf,average='micro'))
# 误差评估

from sklearn.metrics import confusion_matrix

# svc
C=confusion_matrix(y_test, y_svc)
# df=pd.DataFrame(C,index=["财务造假", "财务不造假"],columns=["财务造假", "财务不造假"])
# sns.heatmap(df,annot=True)
plt.matshow(C, cmap=plt.cm.Greens) 
plt.colorbar()
for i in range(len(C)): 
    for j in range(len(C)):
        plt.annotate(C[i,j], xy=(i, j), horizontalalignment='center', verticalalignment='center')
plt.ylabel('True label')
plt.xlabel('Predicted label') 
plt.title('SVC')
plt.show()

# knn
C=confusion_matrix(y_test, y_knn)
# df=pd.DataFrame(C,index=["财务造假", "财务不造假"],columns=["财务造假", "财务不造假"])
# sns.heatmap(df,annot=True)
plt.matshow(C, cmap=plt.cm.Greens) 
plt.colorbar()
for i in range(len(C)): 
    for j in range(len(C)):
        plt.annotate(C[i,j], xy=(i, j), horizontalalignment='center', verticalalignment='center')
plt.ylabel('True label')
plt.xlabel('Predicted label') 
plt.title('KNN')
plt.show()


# log
C=confusion_matrix(y_test, y_logistic)
# df=pd.DataFrame(C,index=["财务造假", "财务不造假"],columns=["财务造假", "财务不造假"])
# sns.heatmap(df,annot=True)
plt.matshow(C, cmap=plt.cm.Greens) 
plt.colorbar()
for i in range(len(C)): 
    for j in range(len(C)):
        plt.annotate(C[i,j], xy=(i, j), horizontalalignment='center', verticalalignment='center')
plt.ylabel('True label')
plt.xlabel('Predicted label') 
plt.title('Logistic')
plt.show()


# dt
C=confusion_matrix(y_test, y_dt)
# df=pd.DataFrame(C,index=["财务造假", "财务不造假"],columns=["财务造假", "财务不造假"])
# sns.heatmap(df,annot=True)
plt.matshow(C, cmap=plt.cm.Greens) 
plt.colorbar()
for i in range(len(C)): 
    for j in range(len(C)):
        plt.annotate(C[i,j], xy=(i, j), horizontalalignment='center', verticalalignment='center')
plt.ylabel('True label')
plt.xlabel('Predicted label') 
plt.title('dt')
plt.show()



# nn
C=confusion_matrix(y_test, y_nn)
# df=pd.DataFrame(C,index=["财务造假", "财务不造假"],columns=["财务造假", "财务不造假"])
# sns.heatmap(df,annot=True)
plt.matshow(C, cmap=plt.cm.Greens) 
plt.colorbar()
for i in range(len(C)): 
    for j in range(len(C)):
        plt.annotate(C[i,j], xy=(i, j), horizontalalignment='center', verticalalignment='center')
plt.ylabel('True label')
plt.xlabel('Predicted label') 
plt.title('nn')
plt.show()