软件技术主要学什么课程,公司seo是什么职位,网站建设移交内容,wordpress 固定链接 分类在本项目中#xff0c;通过数据科学和AI的方法#xff0c;分析挖掘人力资源流失问题#xff0c;并基于机器学习构建解决问题的方法#xff0c;并且#xff0c;我们通过对AI模型的反向解释#xff0c;可以深入理解导致人员流失的主要因素#xff0c;HR部门也可以根据分析…在本项目中通过数据科学和AI的方法分析挖掘人力资源流失问题并基于机器学习构建解决问题的方法并且我们通过对AI模型的反向解释可以深入理解导致人员流失的主要因素HR部门也可以根据分析做出正确的决定。
探索性数据分析 ##1.数据加载
import pandas as pd
import seaborn as sns
data pd.read_csv(../data/train.csv)
#分析建模查看数据情况1.数据包含数值型和类别型
data查看数据基本信息
#字段类型缺失情况 data.info() data.info() 来获取数据的信息包括总行数样本数和总列数字段数、变量的数据类型、数据集中非缺失的数量以及内存使用情况。 从数据集的信息可以看出一共有31 个特征Attrition 是目标字段23个变量是整数类型变量8个是对象类型变量。
2.数据基本分析
#数据无缺失值查看数据分布
data.describe()跑baseline模型使用不同的分类算法
对特征不进行处理
# 选出数值型特征
numerical_feat data.select_dtypes(include[int64])
numerical_feat
# 切分特征和标签
X numerical_feat.drop([Attrition],axis1)
Y numerical_feat.Attrition
# 特征幅度缩放
from sklearn.preprocessing import MinMaxScaler
scaler MinMaxScaler()
x_scaled scaler.fit_transform(X)
x_scaled pd.DataFrame(x_scaled, columnsX.columns)
x_scaled
# 第一次跑模型
## 训练集测试集切分
from sklearn.model_selection import train_test_split
from sklearn.model_selection import train_test_split
from sklearn.metrics import roc_auc_score
from sklearn.tree import DecisionTreeClassifier
from sklearn.linear_model import LogisticRegression
from sklearn.ensemble import RandomForestClassifier
from sklearn.ensemble import GradientBoostingClassifier
import xgboost as xgb
import lightgbm as lgbX_train, X_test, Y_train, Y_test train_test_split(x_scaled,Y,test_size0.3,random_state1)
# 决策树
dt_clf DecisionTreeClassifier()
dt_clf.fit(X_train, Y_train)
dt_auc roc_auc_score(Y_test, dt_clf.predict_proba(X_test)[:, 1])# 逻辑回归
lr_clf LogisticRegression()
lr_clf.fit(X_train, Y_train)
lr_auc roc_auc_score(Y_test, lr_clf.predict_proba(X_test)[:, 1])# 随机森林
rf_clf RandomForestClassifier()
rf_clf.fit(X_train, Y_train)
rf_auc roc_auc_score(Y_test, rf_clf.predict_proba(X_test)[:, 1])# 集成学习 - 梯度提升
gb_clf GradientBoostingClassifier()
gb_clf.fit(X_train, Y_train)
gb_auc roc_auc_score(Y_test, gb_clf.predict_proba(X_test)[:, 1])#XGBoost
xgb_clf xgb.XGBClassifier(eval_metricauc)
xgb_clf.fit(X_train, Y_train)
xgb_auc roc_auc_score(Y_test, xgb_clf.predict_proba(X_test)[:, 1])#LightGBM
lgb_clf lgb.LGBMClassifier()
lgb_clf.fit(X_train, Y_train)
lgb_auc roc_auc_score(Y_test, lgb_clf.predict_proba(X_test)[:, 1])# 打印AUC值
print(fDecision Tree AUC: {dt_auc})
print(fLogistic Regression AUC: {lr_auc})
print(fRandom Forest AUC: {rf_auc})
print(fGradient Boosting AUC: {gb_auc})
print(fXGBoost AUC: {xgb_auc})
print(fLightGBM AUC: {lgb_auc}) 3.特征工程
人才流失中更多的是做特征选择
尝试编码
# 按照出差的频度进行编码
data.BusinessTravel data.BusinessTravel.replace({Non-Travel:0,Travel_Rarely:1,Travel_Frequently:2})# 性别与overtime编码
data.Gender data.Gender.replace({Male:1,Female:0})
data.OverTime data.OverTime.replace({Yes:1,No:0})
data.Over18 data.Over18.replace({Y:1,N:0})
# 独热向量编码
new_df pd.get_dummies(datadata,columns[Department,EducationField,JobRole, MaritalStatus])
new_df# 切分特征和标签
X new_df.drop([Attrition],axis1)
Y new_df.Attrition
# 特征幅度缩放
from sklearn.preprocessing import MinMaxScaler
scaler MinMaxScaler()
x_scaled scaler.fit_transform(X)
x_scaled pd.DataFrame(x_scaled, columnsX.columns)
# 决策树
dt_clf DecisionTreeClassifier()
dt_clf.fit(X_train, Y_train)
dt_auc roc_auc_score(Y_test, dt_clf.predict_proba(X_test)[:, 1])# 逻辑回归
lr_clf LogisticRegression()
lr_clf.fit(X_train, Y_train)
lr_auc roc_auc_score(Y_test, lr_clf.predict_proba(X_test)[:, 1])# 随机森林
rf_clf RandomForestClassifier()
rf_clf.fit(X_train, Y_train)
rf_auc roc_auc_score(Y_test, rf_clf.predict_proba(X_test)[:, 1])# 集成学习 - 梯度提升
gb_clf GradientBoostingClassifier()
gb_clf.fit(X_train, Y_train)
gb_auc roc_auc_score(Y_test, gb_clf.predict_proba(X_test)[:, 1])#XGBoost
xgb_clf xgb.XGBClassifier(eval_metricauc)
xgb_clf.fit(X_train, Y_train)
xgb_auc roc_auc_score(Y_test, xgb_clf.predict_proba(X_test)[:, 1])#LightGBM
lgb_clf lgb.LGBMClassifier()
lgb_clf.fit(X_train, Y_train)
lgb_auc roc_auc_score(Y_test, lgb_clf.predict_proba(X_test)[:, 1])# 打印AUC值
print(fDecision Tree AUC: {dt_auc})
print(fLogistic Regression AUC: {lr_auc})
print(fRandom Forest AUC: {rf_auc})
print(fGradient Boosting AUC: {gb_auc})
print(fXGBoost AUC: {xgb_auc})
print(fLightGBM AUC: {lgb_auc}) 并没有明显提高
特征筛选选出对模型贡献度大的特征
## 训练集测试集切分
from sklearn.model_selection import train_test_split
from sklearn.feature_selection import mutual_info_classif
X_train, X_test, Y_train, Y_test train_test_split(x_scaled,Y,test_size0.3,random_state1)
mutual_info pd.Series(mutual_info)
mutual_info.index X_train.columns
mutual_info.sort_values(ascendingFalse)plt.title(Feature Importance,fontsize20)
mutual_info.sort_values().plot(kindbarh,figsize(12,9),colorr)
plt.show()剔除无效特征后18位
sorted_mutual_info mutual_info.sort_values(ascendingFalse)
# 获取互信息值最低的18个特征的索引列名
least_important_feature_indices sorted_mutual_info.tail(18).index# 从new_df中删除这些特征
new_df new_df.drop(columnsleast_important_feature_indices)
new_df # 切分特征和标签
X new_df.drop([Attrition],axis1)
Y new_df.Attrition
# 特征幅度缩放
from sklearn.preprocessing import MinMaxScaler
scaler MinMaxScaler()
x_scaled scaler.fit_transform(X)
x_scaled pd.DataFrame(x_scaled, columnsX.columns)
## 训练集测试集切分
from sklearn.model_selection import train_test_split
from sklearn.feature_selection import mutual_info_classif
X_train, X_test, Y_train, Y_test train_test_split(x_scaled,Y,test_size0.3,random_state1)
# 定义模型列表
models [(Decision Tree, DecisionTreeClassifier()),(Logistic Regression, LogisticRegression()),(Random Forest, RandomForestClassifier()),(Gradient Boosting, GradientBoostingClassifier()),(XGBoost, xgb.XGBClassifier(eval_metricauc)),(LightGBM, lgb.LGBMClassifier())
]# 训练模型并计算AUC
for name, model in models:model.fit(X_train, Y_train)pred_proba model.predict_proba(X_test)[:, 1]auc roc_auc_score(Y_test, pred_proba)print(f{name} AUC: {auc})有了明显提高
做一些SMOTE
# SMOTE处理类别不均衡
from imblearn.over_sampling import SMOTE
sm SMOTE(sampling_strategyminority)
x,y sm.fit_resample(X,Y)
# 过采样之后的比例
sns.countplot(datanew_df,xy,paletteSet1)
plt.show()
print(y.value_counts()) # 特征幅度缩放
from sklearn.preprocessing import MinMaxScaler
scaler MinMaxScaler()
x_scaled scaler.fit_transform(x)
x_scaled pd.DataFrame(x_scaled, columnsx.columns)## 训练集测试集切分
from sklearn.model_selection import train_test_split
from sklearn.feature_selection import mutual_info_classif
X_train, X_test, Y_train, Y_test train_test_split(x_scaled,y,test_size0.3,random_state1)
# 定义模型列表
models [(Decision Tree, DecisionTreeClassifier()),(Logistic Regression, LogisticRegression()),(Random Forest, RandomForestClassifier()),(Gradient Boosting, GradientBoostingClassifier()),(XGBoost, xgb.XGBClassifier(eval_metricauc)),(LightGBM, lgb.LGBMClassifier())
]# 训练模型并计算AUC
for name, model in models:model.fit(X_train, Y_train)pred_proba model.predict_proba(X_test)[:, 1]auc roc_auc_score(Y_test, pred_proba)print(f{name} AUC: {auc})模型有了大幅度提高
LOF
from pyod.models.lof import LOF
train new_df.copy()
val new_df.copy()
#创建LOF对象
clf LOF(n_neighbors20, algorithmauto)
# 切分特征和标签
X train.drop([Attrition],axis1)#无监督学习算法因此没有y不需要传入y
clf.fit(X)#模型预测
train[out_pred] clf.predict_proba(X)[:,1]
#随机给的一个93%分数数的一个参考值93%是随便给的不宜太小
#判断依据只要小于93%分位数的值就说明这个样本是正常数据如果大于93%分位数的值则说明是异常数据
key train[out_pred].quantile(0.93)
# Attrition 是目标变量列我们不想将其包括在特征列表中
excluded_columns [Attrition]
# 获取所有列名并将排除列从列表中移除
feature_lst [col for col in new_df.columns.tolist() if col not in excluded_columns]
#获取用于模型训练的特征列
x train[train[out_pred] key][feature_lst]
y train[train[out_pred] key][Attrition]#准备验证集的x和y
x train[feature_lst]
y train[Attrition]
val_x val[feature_lst]
val_y val[Attrition]
#模型训练
lr_model LogisticRegression(C0.1,class_weightbalanced)
lr_model.fit(x,y)
from sklearn.metrics import roc_curve#模型预测和画图
y_pred lr_model.predict_proba(x)[:,1]
fpr_lr_train,tpr_lr_train,_ roc_curve(y,y_pred)
train_ks abs(fpr_lr_train - tpr_lr_train).max()
print(train_ks : ,train_ks)y_pred lr_model.predict_proba(val_x)[:,1]
fpr_lr,tpr_lr,_ roc_curve(val_y,y_pred)
val_ks abs(fpr_lr - tpr_lr).max()
print(val_ks : ,val_ks)from matplotlib import pyplot as plt
plt.plot(fpr_lr_train,tpr_lr_train,label train LR)
plt.plot(fpr_lr,tpr_lr,label evl LR)
plt.plot([0,1],[0,1],k--)
plt.xlabel(False positive rate)
plt.ylabel(True positive rate)
plt.title(ROC Curve)
plt.legend(loc best)
plt.show() 交叉验证和超参数调优
网格搜索模型针对具有一定范围值的超参数网格进行评估尝试参数值的每种组合并实验以找到最佳超参数计算成本很高。随机搜索这种方法评估模型的超参数值的随机组合以找到最佳参数计算成本低于网格搜索。 from sklearn.metrics import roc_auc_score
from sklearn.model_selection import cross_val_score# 定义模型列表
models [(Decision Tree, DecisionTreeClassifier()),(Logistic Regression, LogisticRegression()),(Random Forest, RandomForestClassifier()),(Gradient Boosting, GradientBoostingClassifier()),(XGBoost, xgb.XGBClassifier(eval_metricauc)),(LightGBM, lgb.LGBMClassifier())
]# X_train, Y_train, X_test, Y_test是已经准备好的数据集
# X_scaled是经过标准化的特征数据集# 训练模型并计算AUC
for name, model in models:model.fit(X_train, Y_train)pred_proba model.predict_proba(X_test)[:, 1]auc roc_auc_score(Y_test, pred_proba)print(f{name} AUC: {auc})# 使用交叉验证查看得分
for name, model in models:print(******, name, ******)cv_scores cross_val_score(model, x_scaled, y, cv5, scoringroc_auc) # 使用roc_auc作为评分标准cv_mean cv_scores.mean()print(fCross-validated AUC mean score: {cv_mean}) from sklearn.model_selection import GridSearchCV
from sklearn.metrics import make_scorer, roc_auc_score# 定义模型列表
models [(Decision Tree, DecisionTreeClassifier(), {max_depth: [3, 5, 10]}),(Logistic Regression, LogisticRegression(), {C: [0.1, 1, 10]}),(Random Forest, RandomForestClassifier(), {n_estimators: [10, 50, 100]}),(Gradient Boosting, GradientBoostingClassifier(), {n_estimators: [50, 100, 200]}),(XGBoost, xgb.XGBClassifier(eval_metricauc), {n_estimators: [50, 100, 200]}),(LightGBM, lgb.LGBMClassifier(), {n_estimators: [50, 100, 200]})
]# 使用网格搜索进行交叉验证
for name, model, params in models:print(fGrid searching {name}...)grid_search GridSearchCV(model, param_gridparams, cv5, scoringroc_auc)grid_search.fit(X_scaled, y)print(fBest parameters for {name}: {grid_search.best_params_})print(fCross-validated AUC mean score for {name}: {grid_search.best_score_})
