大家好,登顶我是信贷Peter~
本文是一个基于kaggle机器学习实战案例:基于机器学习的信贷违约预测实战,采用了多种模型,违约最终结果随机森林模型排名第一。预测
主要内容包含:
In [1]:
import numpy as npimport pandas as pdimport matplotlib.pyplot as plt%matplotlib inlineimport missingno as msoimport seaborn as snsimport warningsimport osimport scipyfrom scipy import statsfrom scipy.stats import pearsonrfrom scipy.stats import ttest_indfrom sklearn.metrics import classification_reportfrom sklearn.metrics import confusion_matrixfrom sklearn.preprocessing import MinMaxScalerfrom sklearn.model_selection import train_test_splitfrom imblearn.over_sampling import SMOTEfrom sklearn.linear_model import LogisticRegressionfrom sklearn.neighbors import KNeighborsClassifierfrom sklearn.svm import SVCfrom sklearn.naive_bayes import CategoricalNBfrom sklearn.naive_bayes import GaussianNBfrom sklearn.tree import DecisionTreeClassifierfrom sklearn.ensemble import RandomForestClassifierfrom sklearn.ensemble import GradientBoostingClassifierfrom xgboost import XGBClassifierfrom sklearn.model_selection import GridSearchCV,建模 RandomizedSearchCVfrom sklearn.metrics import accuracy_scoreimport warningswarnings.filterwarnings("ignore")
In [2]:
df = pd.read_csv("data.csv")df.head()
Out[2]:
In [3]:
# 整体的数据量df.shape
Out[3]:
(614, 13)
In [4]:
# 全部字段df.columns
Out[4]:
Index(['Loan_ID', 'Gender', 'Married', 'Dependents', 'Education', 'Self_Employed', 'ApplicantIncome', 'CoapplicantIncome', 'LoanAmount', 'Loan_Amount_Term', 'Credit_History', 'Property_Area', 'Loan_Status'], dtype='object')
可以看到主要是贷款人ID、性别、随机森林是登顶否结婚、工作部门、信贷受教育程度等字段信息
In [5]:
# 查看字段类型df.dtypes
Out[5]:
Loan_ID objectGender objectMarried objectDependents objectEducation objectSelf_Employed objectApplicantIncome int64CoapplicantIncome float64LoanAmount float64Loan_Amount_Term float64Credit_History float64Property_Area objectLoan_Status objectdtype: object
In [6]:
# 查看描述统计信息df.describe()
Out[6]:
ApplicantIncome | CoapplicantIncome | LoanAmount | Loan_Amount_Term | Credit_History | |
count | 614.000000 | 614.000000 | 592.000000 | 600.00000 | 564.000000 |
mean | 5403.459283 | 1621.245798 | 146.412162 | 342.00000 | 0.842199 |
std | 6109.041673 | 2926.248369 | 85.587325 | 65.12041 | 0.364878 |
min | 150.000000 | 0.000000 | 9.000000 | 12.00000 | 0.000000 |
25% | 2877.500000 | 0.000000 | 100.000000 | 360.00000 | 1.000000 |
50% | 3812.500000 | 1188.500000 | 128.000000 | 360.00000 | 1.000000 |
75% | 5795.000000 | 2297.250000 | 168.000000 | 360.00000 | 1.000000 |
max | 81000.000000 | 41667.000000 | 700.000000 | 480.00000 | 1.000000 |
In [7]:
df.isnull().sum()
Out[7]:
Loan_ID 0Gender 13Married 3Dependents 15Education 0Self_Employed 32ApplicantIncome 0CoapplicantIncome 0LoanAmount 22Loan_Amount_Term 14Credit_History 50Property_Area 0Loan_Status 0dtype: int64
可以看到部分字段存在缺失值
In [8]:
mso.bar(df,color="blue")plt.show()
图片
后面会针对缺失值进行填充处理。数据探索EDA
In [9]:
df.Loan_ID.value_counts(dropna=False)
Out[9]:
LP001002 1LP002328 1LP002305 1LP002308 1LP002314 1 ..LP001692 1LP001693 1LP001698 1LP001699 1LP002990 1Name: Loan_ID,预测 Length: 614, dtype: int64
可以看到每个Loan_ID刚好一条记录
In [10]:
df.Gender.value_counts(dropna=False)
Out[10]:
Male 489Female 112NaN 13Name: Gender, dtype: int64
In [11]:
sns.countplot(x="Gender", data=df, palette="hls")plt.show()
图片
针对不同性别做缺失值处理:
In [12]:
countMale = len(df[df.Gender == 'Male']) # 男性数据countFemale = len(df[df.Gender == 'Female']) # 女性数据countNull = len(df[df.Gender.isnull()]) # 缺失值数量
In [13]:
print("Percentage of Male: { :.2f}%".format((countMale / (len(df.Gender)) * 100)))print("Percentage of Female: { :.2f}%".format((countFemale / (len(df.Gender)) * 100)))print("Missing values percentage: { :.2f}%".format((countNull / (len(df.Gender)) * 100)))Percentage of Male: 79.64%Percentage of Female: 18.24%Missing values percentage: 2.12%
In [14]:
df.Married.value_counts(dropna=False)
Out[14]:
Yes 398No 213NaN 3Name: Married, dtype: int64
In [15]:
sns.countplot(x="Married", data=df, palette="Paired")plt.show()
图片
是否结婚的人群对比:
In [16]:
countMarried = len(df[df.Married == 'Yes'])countNotMarried = len(df[df.Married == 'No'])countNull = len(df[df.Married.isnull()])
In [17]:
print("Percentage of married: { :.2f}%".format((countMarried / (len(df.Married))*100)))print("Percentage of Not married applicant: { :.2f}%".format((countNotMarried / (len(df.Married))*100)))print("Missing values percentage: { :.2f}%".format((countNull / (len(df.Married))*100)))Percentage of married: 64.82%Percentage of Not married applicant: 34.69%Missing values percentage: 0.49%
In [18]:
df.Education.value_counts(dropna=False)
Out[18]:
Graduate 480Not Graduate 134Name: Education, dtype: int64
In [19]:
sns.countplot(x="Education", data=df, palette="rocket")plt.show()
图片
不同受教育程度的人群对比:
In [20]:
countGraduate = len(df[df.Education == 'Graduate'])countNotGraduate = len(df[df.Education == 'Not Graduate'])countNull = len(df[df.Education.isnull()])print("Percentage of graduate applicant: { :.2f}%".format((countGraduate / (len(df.Education))*100)))print("Percentage of Not graduate applicant: { :.2f}%".format((countNotGraduate / (len(df.Education))*100)))print("Missing percentage: { :.2f}%".format((countNull / (len(df.Education))*100)))Percentage of graduate applicant: 78.18%Percentage of Not graduate applicant: 21.82%Missing percentage: 0.00%
In [21]:
df.Self_Employed.value_counts(dropna=False)
Out[21]:
No 500Yes 82NaN 32Name: Self_Employed, dtype: int64
In [22]:
sns.countplot(x="Self_Employed", data=df, palette="crest")plt.show()
图片
是否为自聘员工对比:
In [23]:
countNo = len(df[df.Self_Employed == 'No'])countYes = len(df[df.Self_Employed == 'Yes'])countNull = len(df[df.Self_Employed.isnull()])print("Percentage of Not self employed: { :.2f}%".format((countNo / (len(df.Self_Employed))*100)))print("Percentage of self employed: { :.2f}%".format((countYes / (len(df.Self_Employed))*100)))print("Missing values percentage: { :.2f}%".format((countNull / (len(df.Self_Employed))*100)))Percentage of Not self employed: 81.43%Percentage of self employed: 13.36%Missing values percentage: 5.21%
In [24]:
df.Credit_History.value_counts(dropna=False)
Out[24]:
1.0 4750.0 89NaN 50Name: Credit_History, dtype: int64
In [25]:
sns.countplot(x="Credit_History", data=df, palette="viridis")plt.show()
图片
是否有信用卡历史的人群对比:
In [26]:
count1 = len(df[df.Credit_History == 1])count0 = len(df[df.Credit_History == 0])countNull = len(df[df.Credit_History.isnull()])
In [27]:
print("Percentage of Good credit history: { :.2f}%".format((count1 / (len(df.Credit_History))*100)))print("Percentage of Bad credit history: { :.2f}%".format((count0 / (len(df.Credit_History))*100)))print("Missing values percentage: { :.2f}%".format((countNull / (len(df.Credit_History))*100)))Percentage of Good credit history: 77.36%Percentage of Bad credit history: 14.50%Missing values percentage: 8.14%
In [28]:
df.Property_Area.value_counts(dropna=False)
Out[28]:
Semiurban 233Urban 202Rural 179Name: Property_Area, dtype: int64
In [29]:
sns.countplot(x="Property_Area", data=df, palette="cubehelix")plt.show()
图片
不同地区的人群对比:
In [30]:
countUrban = len(df[df.Property_Area == 'Urban'])countRural = len(df[df.Property_Area == 'Rural'])countSemiurban = len(df[df.Property_Area == 'Semiurban'])countNull = len(df[df.Property_Area.isnull()])
In [31]:
print("Percentage of Urban: { :.2f}%".format((countUrban / (len(df.Property_Area))*100)))print("Percentage of Rural: { :.2f}%".format((countRural / (len(df.Property_Area))*100)))print("Percentage of Semiurban: { :.2f}%".format((countSemiurban / (len(df.Property_Area))*100)))print("Missing values percentage: { :.2f}%".format((countNull / (len(df.Property_Area))*100)))Percentage of Urban: 32.90%Percentage of Rural: 29.15%Percentage of Semiurban: 37.95%Missing values percentage: 0.00%
这个字段在3个不同的取值下分布是均匀的,而且没有缺失值
In [32]:
df.Loan_Status.value_counts(dropna=False)
Out[32]:
Y 422N 192Name: Loan_Status, dtype: int64
In [33]:
sns.countplot(x="Loan_Status", data=df, palette="YlOrBr")plt.show()
图片
是否贷款的人群占比对比:
In [34]:
countY = len(df[df.Loan_Status == 'Y'])countN = len(df[df.Loan_Status == 'N'])countNull = len(df[df.Loan_Status.isnull()])print("Percentage of Approved: { :.2f}%".format((countY / (len(df.Loan_Status))*100)))print("Percentage of Rejected: { :.2f}%".format((countN / (len(df.Loan_Status))*100)))print("Missing values percentage: { :.2f}%".format((countNull / (len(df.Loan_Status))*100)))Percentage of Approved: 68.73%Percentage of Rejected: 31.27%Missing values percentage: 0.00%
In [35]:
df.Loan_Amount_Term.value_counts(dropna=False)
Out[35]:
360.0 512180.0 44480.0 15NaN 14300.0 13240.0 484.0 4120.0 360.0 236.0 212.0 1Name: Loan_Amount_Term, dtype: int64
In [36]:
sns.countplot(x="Loan_Amount_Term", data=df, palette="rocket")plt.show()
图片
不同贷款周期的人群对比:
In [37]:
count12 = len(df[df.Loan_Amount_Term == 12.0])count36 = len(df[df.Loan_Amount_Term == 36.0])count60 = len(df[df.Loan_Amount_Term == 60.0])count84 = len(df[df.Loan_Amount_Term == 84.0])count120 = len(df[df.Loan_Amount_Term == 120.0])count180 = len(df[df.Loan_Amount_Term == 180.0])count240 = len(df[df.Loan_Amount_Term == 240.0])count300 = len(df[df.Loan_Amount_Term == 300.0])count360 = len(df[df.Loan_Amount_Term == 360.0])count480 = len(df[df.Loan_Amount_Term == 480.0])countNull = len(df[df.Loan_Amount_Term.isnull()])print("Percentage of 12: { :.2f}%".format((count12 / (len(df.Loan_Amount_Term))*100)))print("Percentage of 36: { :.2f}%".format((count36 / (len(df.Loan_Amount_Term))*100)))print("Percentage of 60: { :.2f}%".format((count60 / (len(df.Loan_Amount_Term))*100)))print("Percentage of 84: { :.2f}%".format((count84 / (len(df.Loan_Amount_Term))*100)))print("Percentage of 120: { :.2f}%".format((count120 / (len(df.Loan_Amount_Term))*100)))print("Percentage of 180: { :.2f}%".format((count180 / (len(df.Loan_Amount_Term))*100)))print("Percentage of 240: { :.2f}%".format((count240 / (len(df.Loan_Amount_Term))*100)))print("Percentage of 300: { :.2f}%".format((count300 / (len(df.Loan_Amount_Term))*100)))print("Percentage of 360: { :.2f}%".format((count360 / (len(df.Loan_Amount_Term))*100)))print("Percentage of 480: { :.2f}%".format((count480 / (len(df.Loan_Amount_Term))*100)))print("Missing values percentage: { :.2f}%".format((countNull / (len(df.Loan_Amount_Term))*100)))Percentage of 12: 0.16%Percentage of 36: 0.33%Percentage of 60: 0.33%Percentage of 84: 0.65%Percentage of 120: 0.49%Percentage of 180: 7.17%Percentage of 240: 0.65%Percentage of 300: 2.12%Percentage of 360: 83.39%Percentage of 480: 2.44%Missing values percentage: 2.28%
In [38]:
df.select_dtypes(exclude=["object"]).columns
Out[38]:
Index(['ApplicantIncome', 'CoapplicantIncome', 'LoanAmount', 'Loan_Amount_Term', 'Credit_History'], dtype='object')
In [39]:
df[['ApplicantIncome','CoapplicantIncome','LoanAmount']].describe()
Out[39]:
ApplicantIncome | CoapplicantIncome | LoanAmount | |
count | 614.000000 | 614.000000 | 592.000000 |
mean | 5403.459283 | 1621.245798 | 146.412162 |
std | 6109.041673 | 2926.248369 | 85.587325 |
min | 150.000000 | 0.000000 | 9.000000 |
25% | 2877.500000 | 0.000000 | 100.000000 |
50% | 3812.500000 | 1188.500000 | 128.000000 |
75% | 5795.000000 | 2297.250000 | 168.000000 |
max | 81000.000000 | 41667.000000 | 700.000000 |
In [40]:
sns.set(style="darkgrid")fig, axs = plt.subplots(2, 2, figsize=(10, 8))sns.histplot(data=df, x="ApplicantIncome", kde=True, ax=axs[0, 0], color='green')sns.histplot(data=df, x="CoapplicantIncome", kde=True, ax=axs[0, 1], color='skyblue')sns.histplot(data=df, x="LoanAmount", kde=True, ax=axs[1, 0], color='orange');
图片
可以看到这3个字段呈现一定的偏态,后面会做数据转换处理。
In [41]:
sns.set(style="darkgrid")fig, axs1 = plt.subplots(2, 2, figsize=(10, 10))sns.violinplot(data=df, y="ApplicantIncome", ax=axs1[0, 0], color='green')sns.violinplot(data=df, y="CoapplicantIncome", ax=axs1[0, 1], color='skyblue')sns.violinplot(data=df, y="LoanAmount", ax=axs1[1, 0], color='orange');
图片
分类型变量主要是基于统计分析查看分布情况:
In [42]:
pd.crosstab(df.Gender,df.Married).plot(kind="bar", stacked=True, figsize=(5,5), color=['#f64f59','#12c2e9'])plt.title('Gender vs Married')plt.xlabel('Gender')plt.ylabel('Frequency')plt.xticks(rotatinotallow=0)plt.show()
图片
好坏信用卡历史对比:
In [43]:
pd.crosstab(df.Self_Employed,df.Credit_History).plot(kind="bar", stacked=True, figsize=(5,5), color=['#544a7d','#ffd452'])plt.title('Self Employed vs Credit History')plt.xlabel('Self Employed')plt.ylabel('Frequency')plt.legend(["Bad Credit", "Good Credit"])plt.xticks(rotatinotallow=0)plt.show()
图片
不同地区的人群是否贷款对比:
In [44]:
pd.crosstab(df.Property_Area,df.Loan_Status).plot(kind="bar", stacked=True, figsize=(5,5), color=['#333333','#dd1818'])plt.title('Property Area vs Loan Status')plt.xlabel('Property Area')plt.ylabel('Frequency')plt.xticks(rotatinotallow=0)plt.show()
图片
In [45]:
sns.boxplot(x="Loan_Status", y="ApplicantIncome", data=df, palette="mako");
图片
sns.boxplot(x="CoapplicantIncome", y="Loan_Status", data=df, palette="rocket");
图片
sns.boxplot(x="Loan_Status", y="LoanAmount", data=df, palette="YlOrBr");
图片
In [48]:
df.plot(x='ApplicantIncome', y='CoapplicantIncome', style='o') plt.title('Applicant Income - Co Applicant Income') plt.xlabel('ApplicantIncome')plt.ylabel('CoapplicantIncome') plt.show()
图片
相关性计算:
In [49]:
print('Pearson correlation:', df['ApplicantIncome'].corr(df['CoapplicantIncome']))print('T Test and P value: \n', stats.ttest_ind(df['ApplicantIncome'], df['CoapplicantIncome']))Pearson correlation: -0.11660458122889966T Test and P value: Ttest_indResult(statistic=13.835753259915661, pvalue=1.4609839484240346e-40)
In [50]:
plt.figure(figsize=(10,7))sns.heatmap(df.corr(), annot=True, cmap='inferno')plt.show()
图片
In [51]:
df.drop("Loan_ID",axis=1, inplace=True)
In [52]:
df.isnull().sum()
Out[52]:
Gender 13Married 3Dependents 15Education 0Self_Employed 32ApplicantIncome 0CoapplicantIncome 0LoanAmount 22Loan_Amount_Term 14Credit_History 50Property_Area 0Loan_Status 0dtype: int64
In [53]:
df.dtypes
Out[53]:
Gender objectMarried objectDependents objectEducation objectSelf_Employed objectApplicantIncome int64CoapplicantIncome float64LoanAmount float64Loan_Amount_Term float64Credit_History float64Property_Area objectLoan_Status objectdtype: object
In [54]:
df["Credit_History"].value_counts()
Out[54]:
1.0 4750.0 89Name: Credit_History, dtype: int64
In [55]:
df["Loan_Amount_Term"].value_counts()
Out[55]:
360.0 512180.0 44480.0 15300.0 13240.0 484.0 4120.0 360.0 236.0 212.0 1Name: Loan_Amount_Term, dtype: int64
针对分类型变量的缺失值,我们使用众数mode进行填充:
In [56]:
df['Gender'].fillna(df['Gender'].mode()[0],inplace=True)df['Married'].fillna(df['Married'].mode()[0],inplace=True)df['Dependents'].fillna(df['Dependents'].mode()[0],inplace=True)df['Self_Employed'].fillna(df['Self_Employed'].mode()[0],inplace=True)# 信用卡历史记录 0-bad credit 1-good credit history df['Credit_History'].fillna(df['Credit_History'].mode()[0],inplace=True)# 还款周期(天)df['Loan_Amount_Term'].fillna(df['Loan_Amount_Term'].mode()[0],inplace=True)
数值型变量的缺失值使用均值mean进行填充
In [57]:
df['LoanAmount'].fillna(df['LoanAmount'].mean(),inplace=True) # 贷款金额
In [58]:
df = pd.get_dummies(df)df.head()
处理后的全部字段信息:
In [59]:
df.columns
Out[59]:
Index(['ApplicantIncome', 'CoapplicantIncome', 'LoanAmount', 'Loan_Amount_Term', 'Credit_History', 'Gender_Female', 'Gender_Male', 'Married_No', 'Married_Yes', 'Dependents_0', 'Dependents_1', 'Dependents_2', 'Dependents_3+', 'Education_Graduate', 'Education_Not Graduate', 'Self_Employed_No', 'Self_Employed_Yes', 'Property_Area_Rural', 'Property_Area_Semiurban', 'Property_Area_Urban', 'Loan_Status_N', 'Loan_Status_Y'], dtype='object')
In [60]:
# 删除部分字段df = df.drop(['Gender_Female', 'Married_No', 'Education_Not Graduate','Self_Employed_No', 'Loan_Status_N'], axis = 1)
In [61]:
# 字段重命名new = { 'Gender_Male': 'Gender', 'Married_Yes': 'Married', 'Education_Graduate': 'Education', 'Self_Employed_Yes': 'Self_Employed', 'Loan_Status_Y': 'Loan_Status'} df.rename(columns=new, inplace=True)
以上下4分位数作为临界点:
In [62]:
Q1 = df.quantile(0.25)Q3 = df.quantile(0.75)IQR = Q3 - Q1df = df[~((df < (Q1 - 1.5 * IQR)) | (df > (Q3 + 1.5 * IQR))).any(axis=1)]
对数据开平方做数据转换:np.sqrt
In [63]:
df.ApplicantIncome = np.sqrt(df.ApplicantIncome)df.CoapplicantIncome = np.sqrt(df.CoapplicantIncome)df.LoanAmount = np.sqrt(df.LoanAmount)
再次查看数据分布:
In [64]:
sns.set(style="darkgrid")fig, axs = plt.subplots(2, 2, figsize=(10, 8))sns.histplot(data=df, x="ApplicantIncome", kde=True, ax=axs[0, 0], color='green')sns.histplot(data=df, x="CoapplicantIncome", kde=True, ax=axs[0, 1], color='skyblue')sns.histplot(data=df, x="LoanAmount", kde=True, ax=axs[1, 0], color='orange');
图片
In [65]:
X = df.drop(["Loan_Status"], axis=1)y = df["Loan_Status"]
In [66]:
pd.value_counts(y) # 采样前
Out[66]:
1 1120 24Name: Loan_Status, dtype: int64
In [67]:
X, y = SMOTE().fit_resample(X, y)
In [68]:
pd.value_counts(y) # 采样后
Out[68]:
1 1120 112Name: Loan_Status, dtype: int64
In [69]:
sns.set_theme(style="darkgrid")sns.countplot(y=y, data=df, palette="coolwarm")plt.ylabel('Loan Status')plt.xlabel('Total')plt.show()
图片
In [70]:
mm = MinMaxScaler()X = mm.fit_transform(X)
In [71]:
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = 0.2, random_state = 0)
In [72]:
pd.value_counts(y_train)
Out[72]:
1 900 89Name: Loan_Status, dtype: int64
In [73]:
pd.value_counts(y_test)
Out[73]:
0 231 22Name: Loan_Status, dtype: int64
In [74]:
LRclassifier = LogisticRegression(solver='saga', max_iter=500, random_state=1)LRclassifier.fit(X_train, y_train)# 模型预测y_pred = LRclassifier.predict(X_test)
In [75]:
print(classification_report(y_test, y_pred)) # 分类结果报告 precision recall f1-score support 0 0.86 0.83 0.84 23 1 0.83 0.86 0.84 22 accuracy 0.84 45 macro avg 0.84 0.84 0.84 45weighted avg 0.85 0.84 0.84 45
In [76]:
print(confusion_matrix(y_test, y_pred)) # 混淆矩阵[[19 4] [ 3 19]]
In [77]:
LRAcc = accuracy_score(y_pred,y_test) # 准确率print('LR accuracy: { :.2f}%'.format(LRAcc * 100))LR accuracy: 84.44%
新版本报错解决,参考:https://blog.csdn.net/weixin_51723388/article/details/128577782
在使用KNeighborsClassifier(n_neighbors=5, *, weights='uniform', algorithm='auto', leaf_size=30, p=2, metric='minkowski', metric_params=None, n_jobs=None)时,只有在weights='uniform' 时才会用到stats.mode。其中uniform是均等权重,即邻域中的所有点的权重相等,相当于取众数。可将其改为weights='distance'
In [78]:
score_list = []for i in range(1,21): knn = KNeighborsClassifier(n_neighbors = i,weights='distance') knn.fit(X_train, y_train) score_list.append(knn.score(X_test, y_test)) # 测试集预测得分plt.plot(range(1,21), score_list)plt.xticks(np.arange(1,21,1))plt.xlabel("K value")plt.ylabel("Score")plt.show()KNAcc = max(score_list)print("KNN best accuracy: { :.2f}%".format(KNAcc*100))
KNN best accuracy: 86.67%
图片
In [79]:
svc = SVC(kernel='rbf', max_iter=500)svc.fit(X_train, y_train)y_pred = svc.predict(X_test)print(classification_report(y_test, y_pred))print(confusion_matrix(y_test, y_pred)) precision recall f1-score support 0 0.95 0.78 0.86 23 1 0.81 0.95 0.88 22 accuracy 0.87 45 macro avg 0.88 0.87 0.87 45weighted avg 0.88 0.87 0.87 45[[18 5] [ 1 21]]
In [80]:
SVCAcc = accuracy_score(y_pred,y_test)print('SVC accuracy: { :.2f}%'.format(SVCAcc*100))SVC accuracy: 86.67%
In [81]:
NBclassifier2 = GaussianNB()NBclassifier2.fit(X_train, y_train)y_pred = NBclassifier2.predict(X_test)print(classification_report(y_test, y_pred))print(confusion_matrix(y_test, y_pred)) precision recall f1-score support 0 0.68 0.83 0.75 23 1 0.76 0.59 0.67 22 accuracy 0.71 45 macro avg 0.72 0.71 0.71 45weighted avg 0.72 0.71 0.71 45[[19 4] [ 9 13]]
In [82]:
GNBAcc = accuracy_score(y_pred,y_test)print('Gaussian Naive Bayes accuracy: { :.2f}%'.format(GNBAcc*100))Gaussian Naive Bayes accuracy: 71.11%
In [83]:
scoreListDT = []for i in range(2,21): DTclassifier = DecisionTreeClassifier(max_leaf_nodes=i) DTclassifier.fit(X_train, y_train) scoreListDT.append(DTclassifier.score(X_test, y_test))plt.plot(range(2,21), scoreListDT)plt.xticks(np.arange(2,21,1))plt.xlabel("Leaf")plt.ylabel("Score")plt.show()DTAcc = max(scoreListDT)print("Decision Tree Accuracy: { :.2f}%".format(DTAcc*100))
Decision Tree Accuracy: 84.44%
图片
In [84]:
scoreListRF = []for i in range(2,25): RFclassifier = RandomForestClassifier(n_estimators = 1000, random_state = 1, max_leaf_nodes=i) RFclassifier.fit(X_train, y_train) scoreListRF.append(RFclassifier.score(X_test, y_test))plt.plot(range(2,25), scoreListRF)plt.xticks(np.arange(2,25,1))plt.xlabel("RF Value")plt.ylabel("Score")plt.show()RFAcc = max(scoreListRF)print("Random Forest Accuracy: { :.2f}%".format(RFAcc*100))
Random Forest Accuracy: 91.11%
图片
In [85]:
# 设置参数params={ 'n_estimators':[100,200,300,400,500], 'max_depth':[1,2,3,4,5], 'subsample':[0.5,1], 'max_leaf_nodes':[2,5,10,20,30,40,50]}
In [86]:
# 基于随机搜索查找参数组合GB = RandomizedSearchCV(GradientBoostingClassifier(), params, cv=20)GB.fit(X_train, y_train)
Out[86]:
RandomizedSearchCV(cv=20, estimator=GradientBoostingClassifier(), param_distributinotallow={ 'max_depth': [1, 2, 3, 4, 5], 'max_leaf_nodes': [2, 5, 10, 20, 30, 40, 50], 'n_estimators': [100, 200, 300, 400, 500], 'subsample': [0.5, 1]})
In [87]:
print(GB.best_estimator_)print(GB.best_score_)GradientBoostingClassifier(max_depth=4, max_leaf_nodes=10, n_estimators=500, subsample=1)0.7993055555555555
In [88]:
print(GB.best_params_) # 最佳参数组合{ 'subsample': 1, 'n_estimators': 500, 'max_leaf_nodes': 10, 'max_depth': 4}
In [89]:
GB.best_params_["subsample"]
Out[89]:
1
基于查找到的参数再重新建模:
In [90]:
gbc = GradientBoostingClassifier(subsample=GB.best_params_["subsample"], n_estimators=GB.best_params_["n_estimators"], max_depth=GB.best_params_["max_depth"], max_leaf_nodes=GB.best_params_["max_leaf_nodes"], )gbc.fit(X_train, y_train)y_pred = gbc.predict(X_test)print(classification_report(y_test, y_pred))print(confusion_matrix(y_test, y_pred)) precision recall f1-score support 0 0.78 0.91 0.84 23 1 0.89 0.73 0.80 22 accuracy 0.82 45 macro avg 0.83 0.82 0.82 45weighted avg 0.83 0.82 0.82 45[[21 2] [ 6 16]]
In [91]:
GBAcc = accuracy_score(y_pred,y_test)print('Gradient Boosting accuracy: { :.2f}%'.format(GBAcc*100))Gradient Boosting accuracy: 82.22%
In [92]:
models = pd.DataFrame({ 'Model': ['Logistic Regression', 'K Neighbors', 'Support Vector Machine', 'Gaussian NB', 'Decision Tree', 'Random Forest', 'Gradient Boost'], 'Accuracy': [LRAcc*100, KNAcc*100, SVCAcc*100,GNBAcc*100, DTAcc*100, RFAcc*100, GBAcc*100]})models.sort_values(by='Accuracy', ascending=False)
Out[92]:
Model | Accuracy | |
5 | Random Forest | 91.111111 |
1 | K Neighbors | 86.666667 |
2 | Support Vector Machine | 86.666667 |
0 | Logistic Regression | 84.444444 |
4 | Decision Tree | 84.444444 |
6 | Gradient Boost | 82.222222 |
3 | Gaussian NB | 71.111111 |
(责任编辑:百科)
经典重制《神技盗来》战斗系统情报 10月13日NS/PS4发售
Springboot整合Camunda工作流引擎实现审批流程实例