精品教学案例 | 基于分类算法的肝病诊断

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns
patient = pd.read_csv("./input/indian_liver_patient.csv")
name = ["Age","Gender","TBIL","DBIL","ALP","ALT","AST","TP","ALB","A/G","Dataset"]
patient.columns = name
patient.head()

patient.info()

#Gender特征转换
patient["Gender"].replace("Female",1,inplace = True)
patient["Gender"].replace("Male",0,inplace = True)
patient["Gender"] = patient["Gender"].astype("int8")

#缺失值处理
patient.isnull().sum(axis = 0)

print("The proportion of missing values is %.5f  "%(patient.isnull().sum(axis = 0).max()/patient.shape[0]))  #缺失值占比

patient.dropna(inplace = True)
print("The shape of data is ",patient.shape)

fig,ax = plt.subplots(2,4,figsize = (15,7))
for i in range(1,9):
    sub = int("24"+str(i))
    plt.subplot(sub)
    plt.hist(patient[name[i+1]],bins = 15)
    plt.xlabel(name[i+1],fontsize = 13)
    plt.ylabel("Count",fontsize = 13)  
plt.tight_layout()
plt.show()   

#对数变换
patient[["ALP","ALT","AST"]] = np.log(patient[["ALP","ALT","AST"]])

fig,ax = plt.subplots(1,3,figsize = (12,4))
for i in range(1,4):
    sub = int("13"+str(i))
    plt.subplot(sub)
    plt.hist(patient[name[i+3]],bins = 15)
    plt.xlabel(name[i+3],fontsize = 13)
    plt.ylabel("Count",fontsize = 13)  
plt.tight_layout()
plt.show()   

fig, ax = plt.subplots(figsize = (7,5))
name_list = ['Male','Fmale'] 
gen_count = patient.groupby(['Gender','Dataset']).Age.count()
ill_count = [gen_count[0][1],gen_count[1][1]]  
notill_count = [gen_count[0][2],gen_count[1][2]]  
x = list(range(len(name_list)))  
total_width, n = 0.8, 2  
width = total_width / n  

plt.bar(x,ill_count,width = width,label = 'ill',color = 'lightsalmon')  
for i in range(len(x)):  
    x[i] = x[i] + width  
plt.bar(x,notill_count, width = width, label = 'notill',tick_label = name_list,color = 'lightskyblue')  
plt.legend(loc = "upper right")  
plt.xlabel("Gender",fontsize = 13)
plt.ylabel("Count",fontsize = 13)
plt.title('Patient by gender')
plt.show()  

fig, ax = plt.subplots(figsize = (7,5))
x = [patient[patient["Dataset"] == 1]["Age"],patient[patient["Dataset"] == 2]["Age"]]
bins = range(0, 91, 10)

plt.hist(x, bins = bins, color = ['lightsalmon','lightskyblue'], histtype = "bar", rwidth = 10,stacked = True, label = ["ill", "notill"])
plt.xlabel("Age",fontsize = 13)
plt.ylabel("Count",fontsize = 13)
plt.title("Patient by Age")
plt.legend(loc = "upper right")
plt.show() 

#特征相关系数图
import matplotlib
matplotlib.rcParams['axes.unicode_minus']=False
fig, ax = plt.subplots(figsize = (8,8))
corr=patient.corr()
sns.heatmap(corr,annot = True, vmax = 1,vmin=-1,square = True, cmap = "Blues")
plt.ylabel('Feature',fontsize = 15)
plt.xlabel('Feature',fontsize = 15)
plt.title("The correlation of features",fontsize = 15)
plt.show()

from sklearn.model_selection import train_test_split
from sklearn.metrics import classification_report
from sklearn.metrics import confusion_matrix
from sklearn.metrics import accuracy_score   

##训练集测试集的划分
x = patient.drop("Dataset",axis = 1)
y = patient["Dataset"]
y.replace(2,0,inplace = True)
x_train, x_test, y_train, y_test = train_test_split(x, y,test_size = 0.25, random_state = 2)

#定义一个模型评价函数
def evaluate(method,y_test,y_pred):
    print(classification_report(y_test,y_pred))
    print(confusion_matrix(y_test, y_pred))
    print("The acc of %s is %.5f"%(method,accuracy_score(y_test,y_pred)))

#自定义逐步回归的函数
import statsmodels.api as sm
def stepwise(x,y,alpha_in = 0.1,alpha_out = 0.15):
    '''x为所有可能的特征构成的DataFrame,
      y为响应变量，
      alpha_in为引入特征的显著性水平，
      alpha_out为剔除特征的显著性水平'''
    included = []
    while True:
        changed = False
        excluded = list(set(x.columns)-set(included))
        p_val = pd.Series(index = excluded)
        for new_column in excluded:
            model = sm.Logit(y,sm.add_constant(x[included+[new_column]])).fit()
            p_val[new_column] = model.pvalues[new_column]
        min_pval = p_val.min()
        #forward step
        if min_pval < alpha_in:
            changed = True
            add_feature = p_val.idxmin()
            included.append(add_feature)
            print("Add {} with p_value   {:.6}".format(add_feature,min_pval))
        #backward step
        model = sm.Logit(y,sm.add_constant(x[included])).fit()
        p_val = model.pvalues.iloc[1:]
        max_pval = p_val.max()
        if max_pval > alpha_out:
            changed = True
            drop_feature = p_val.idxmax()
            included.remove(drop_feature)
            print("Drop {} with p_value   {:.6}".format(drop_feature,max_pval))
        if not changed:
            break
    return included

result = stepwise(x_train,y_train) #这里的结果较长，暂不做展示

new_columns = result
model = sm.Logit(y_train,sm.add_constant(x_train[new_columns])).fit()
model.summary2()

y_pred = model.predict(sm.add_constant(x_test[new_columns]))
y_pred1 = [1 if y>0.6 else 0 for y in y_pred]
evaluate("StepLogit",y_test,y_pred1)

## 决策树
from sklearn.tree import DecisionTreeClassifier
dt = DecisionTreeClassifier(criterion = 'gini',max_depth = 20,random_state = 50,class_weight = "balanced").fit(x_train,y_train)
y_pred = dt.predict(x_test)
evaluate("DecisionTree",y_test,y_pred)

## 随机森林
from sklearn.ensemble import RandomForestClassifier
rf = RandomForestClassifier(n_estimators = 100, max_depth = 20,max_features = "auto", random_state = 20).fit(x_train,y_train)
y_pred = rf.predict(x_test)
evaluate("RandomForest",y_test,y_pred)

from sklearn.ensemble import AdaBoostClassifier
ab = AdaBoostClassifier(n_estimators = 200,learning_rate = 1,algorithm = 'SAMME.R', random_state = 20).fit(x_train,y_train)
y_pred = ab.predict(x_test)
evaluate("AdaBoost",y_test,y_pred)

model = [dt,rf,ab]
name = ["Decision Tree","Random Forest","AdaBoost"]
fig,ax = plt.subplots(1,3,figsize = (15,6))
for i in range(3):
    sub = int("13"+str(i+1))
    x = list(patient.columns[:-1])
    y = list(model[i].feature_importances_)
    plt.subplot(sub,sharey = ax[0])
    plt.bar(x,y) 
    plt.xlabel("Feature",fontsize = 13)
    plt.ylabel("Feature_importance",fontsize = 13)
    plt.title("The features' importance of %s "%name[i])  
plt.tight_layout()
plt.show()   

del x_train["Gender"]
del x_test["Gender"]

dt1 = DecisionTreeClassifier(criterion = 'gini',max_depth = 20,random_state = 50,class_weight = "balanced").fit(x_train,y_train)
y_pred = dt1.predict(x_test)
print("-------------The result of DecisionTree-------------\n")
evaluate("DecisionTree",y_test,y_pred)

rf1 = RandomForestClassifier(n_estimators = 100, max_depth = 20,max_features = "auto", random_state = 20).fit(x_train,y_train)
y_pred = rf1.predict(x_test)
print("\n-------------The result of RandomForset-------------\n")
evaluate("RandomForest",y_test,y_pred)

ab1 = AdaBoostClassifier(n_estimators = 200,learning_rate = 1,algorithm = 'SAMME.R', random_state = 20)
ab1.fit(x_train,y_train)
y_pred = ab1.predict(x_test)
print("\n---------------The result of AdaBoost---------------\n")
evaluate("AdaBoost",y_test,y_pred)