Travel Tips
Lorem ipsum dolor sit amet, consectetur adipiscing elit.
上市公司财务预警（神经网络）
上市公司财务预警（神经网络）
业务背景
神经网络理论初探
R语言实现
神经网络优缺点
BP神经网络
R语言代码实现
#设置工作空间
getwd()
setwd("/home/zty/Documents/R/上市公司财务预警")
library(nnet)
#导入数据集
data_bp <- read.csv("shuru.csv", fileEncoding = 'GBK')
head(data_bp)
#提取数据子集
data_bp_sub <- data_bp[,c("Datatype","ST","AvgROE","ROAEBIT","Opeprfrt","Currt","Qckrt","Opeincmgrrt",
"Netprfgrrt","Netassgrrt","Invtrtrrat","ARTrat","Currat",
"Totassrat","Casrtsale","Dbastrt","Ebitada")]
#查看数据
head(data_bp_sub)
table(data_bp_sub$ST)
#数据变换
data_bp_sub <- within (data_bp_sub,{
ST_1 <- NA
ST_1[ST == "是"] <- 1
ST_1[ST == "否"] <- 0})
#查看数据
table(data_bp_sub$ST,data_bp_sub$ST_1)
#构建训练集与测试集
data_bp_sub_train <- data_bp_sub[which(data_bp_sub$Datatype == "训练集"),]
data_bp_sub_test <- data_bp_sub[which(data_bp_sub$Datatype == "测试集"),]
#查看数据类型
str(data_bp_sub_train)
str(data_bp_sub_test)
#改变数据类型
data_bp_sub_train$AvgROE <- as.numeric(data_bp_sub_train$AvgROE)
data_bp_sub_train$Netprfgrrt <- as.numeric(data_bp_sub_train$Netprfgrrt)
data_bp_sub_train$Netassgrrt <- as.numeric(data_bp_sub_train$Netassgrrt)
data_bp_sub_train$Ebitada <- as.numeric(data_bp_sub_train$Ebitada)
data_bp_sub_train$ST_1 <- as.factor(data_bp_sub_train$ST_1)
data_bp_sub_test$AvgROE <- as.numeric(data_bp_sub_test$AvgROE)
data_bp_sub_test$Netprfgrrt <- as.numeric(data_bp_sub_test$Netprfgrrt)
data_bp_sub_test$Netassgrrt <- as.numeric(data_bp_sub_test$Netassgrrt)
data_bp_sub_test$Ebitada <- as.numeric(data_bp_sub_test$Ebitada)
data_bp_sub_test$ST_1 <- as.factor(data_bp_sub_test$ST_1)
# 构建nnet神经网络
model1=nnet(ST_1~AvgROE+ROAEBIT+Opeprfrt+Currt+Qckrt+Opeincmgrrt+Netprfgrrt+Netassgrrt
+Invtrtrrat+ARTrat+Currat+Totassrat+Casrtsale+Dbastrt+Ebitada,
data=data_bp_sub_train,size=6,decay=5e-4,rang = 0.1,maxit=1000)
model1
#构建混淆矩阵
confusion=table(data_bp_sub_test$ST_1,predict(model1,data_bp_sub_test[,3:17],type ="class"))
confusionPython实现
#导入pandas库 import pandas as pd #导入数据集 net = pd.read_csv ( '/media/zty/aae84acb-4d27-47dc-b5a0-93a21d9f7cd9/zty/数据分析文件/262-结实-左手R右手Python数据挖掘机器学习/第五章/神经网络--python实现/shuru.csv',encoding='utf-8') net.head()
#提取数据子集
net_sub = net.loc[:, ['ROAEBIT', 'Dbastrt','Currt','Qckrt','Invtrtrrat','ARTrat','Currat', 'Totassrat','Casrtsale','Opeprfrt','Ebitada','AvgROE','Opeincmgrrt', 'Netprfgrrt','Netassgrrt','ST']] #条件赋值 #net_sub['ST_1'] = net_sub.ST.apply(lambda x: 1 if '是' in x else 0) #类型查看 net_sub.dtypes
#删除不必要的列
net_sub_s = net_sub.drop('ST',axis = 1)
#再查看数据集
net_sub_s.dtypes
#net_sub_s.head()
拆分训练集与测试集
from sklearn.model_selection import train_test_split array = net_sub_s.values array = net_sub_s.values nb_classes = 6 X =array[:,0:15] Y = array[:,14] test_size = 0.30 seed = 4 X_train, X_test, Y_train, Y_test = train_test_split(X, Y, test_size=test_size, random_state=seed) # 一般形式：X_train,X_test, y_train, y_test =cross_validation.train_test_split(train_data,train_target,test_size= , random_state= ) #X_train：所要划分的样本特征集 #train_target：所要划分的样本结果 #test_size：样本占比，如果是整数的话就是样本的数量 #random_state：是随机数的种子。在需要重复试验的时候，保证得到一组一样的随机数。
搭建神经网络模型
#导入人工神经网络库 from sklearn.neural_network import MLPClassifier mlp = MLPClassifier(hidden_layer_sizes= ( 13,13,13 ) ,max_iter=500) mlp
mlp.fit(X_train.astype(int),Y_train.astype(int))
predictions = mlp.predict ( X_test ) from sklearn.metrics import classification_report,confusion_matrix import numpy as np print ( confusion_matrix ( np.rint(Y_test),predictions ) )
#评估具体指标
from sklearn import metrics
print("tes_score:", mlp.score(np.rint(X_test), np.rint(Y_test)))
y_pred = mlp.predict(X_test)
print("查准率:",metrics.precision_score(np.rint(Y_test), np.rint(y_pred),average='weighted'))
print("召回率:",metrics.recall_score(np.rint(Y_test), np.rint(y_pred),average='weighted'))
print("F1_score:",metrics.f1_score(np.rint(Y_test), np.rint(y_pred),average='weighted'))
Sed ac lorem felis. Ut in odio lorem. Quisque magna dui, maximus ut commodo sed, vestibulum ac nibh. Aenean a tortor in sem tempus auctor
Agatha Christie
December 4, 2020 at 3:12 pm
Donec in ullamcorper quam. Aenean vel nibh eu magna gravida fermentum. Praesent eget nisi pulvinar, sollicitudin eros vitae, tristique odio.
Danielle Steel
December 4, 2020 at 3:12 pm
我是 s enim interduante quis metus. Duis porta ornare nulla ut bibendum
Rosie
6 minutes ago