【Python】Risk Factors

kaggle discussion - Adding Risk Factors

import numpy as np # linear algebra
import pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)

import seaborn as sns
import matplotlib.pyplot as plt

Load data and look at distributions of features

train_df = pd.read_csv('/kaggle/input/playground-series-s3e3/train.csv')
train_df['NumCompaniesWorked'] = train_df['NumCompaniesWorked'].replace(0, 1)
train_df.head()

	id	Age	BusinessTravel	DailyRate	Department	DistanceFromHome	Education	EducationField	EmployeeCount	EnvironmentSatisfaction	...	StandardHours	StockOptionLevel	TotalWorkingYears	TrainingTimesLastYear	WorkLifeBalance	YearsAtCompany	YearsInCurrentRole	YearsSinceLastPromotion	YearsWithCurrManager	Attrition
0	0	36	Travel_Frequently	599	Research & Development	24	3	Medical	1	4	...	80	1	10	2	3	10	0	7	8	0
1	1	35	Travel_Rarely	921	Sales	8	3	Other	1	1	...	80	1	4	3	3	4	2	0	3	0
2	2	32	Travel_Rarely	718	Sales	26	3	Marketing	1	3	...	80	2	4	3	3	3	2	1	2	0
3	3	38	Travel_Rarely	1488	Research & Development	2	3	Medical	1	3	...	80	0	15	1	1	6	0	0	2	0
4	4	50	Travel_Rarely	1017	Research & Development	5	4	Medical	1	2	...	80	0	31	0	3	31	14	4	10	1

5 rows × 35 columns

Additional features

train_df['AverageTenure'] = train_df["TotalWorkingYears"] / train_df["NumCompaniesWorked"]
train_df['YearsAboveAvgTenure'] = train_df['YearsAtCompany'] - train_df['AverageTenure']

train_df['JobHopper'] = ((train_df["NumCompaniesWorked"] > 2) & (train_df["AverageTenure"] < 2.0)).astype(int)
train_df['MonthlyIncome_Age'] = train_df['MonthlyIncome'] / train_df['Age']

Visualizing Risk Factors

See How are risk factors determined? and Visualizing risk factors for stroke from the previous competition.

sns.set_style("whitegrid", {"xtick.bottom" : True, "ytick.left" : True})

df_train0 = train_df[train_df["Attrition"] == 0]
df_train1 = train_df[train_df["Attrition"] == 1]
features = ["Age", "HourlyRate", "DistanceFromHome", "AverageTenure", "YearsAboveAvgTenure", "JobHopper",
            "YearsAtCompany", "YearsInCurrentRole", "YearsSinceLastPromotion", "YearsWithCurrManager",
            "MonthlyIncome_Age"]

for feature in features:
    plt.figure(figsize=(12, 6))
    '''
    核密度估计(kernel density estimation)
    
    '''
    sns.kdeplot(df_train0[feature], shade=True, label="0")
    sns.kdeplot(df_train1[feature], shade=True, label="1")
    plt.title("{}".format(feature), fontsize=15)
    plt.ylabel("y")
    plt.xlabel("x")
    plt.tight_layout()
    plt.savefig("distributions_%s.png" % feature, dpi=150)
    plt.legend()
    plt.show()

s

Insights:

• It looks like we can set risk factors for Age and HourlyRate.
• There is a small increased risk with DistanceFromHome.
• There is a risk involved with all of the features YearsAtCompany, YearsInCurrentRole, YearsWithCurrManager, but the last two might just correlate highly to YearsAtCompany.
• A low MonthlyIncome / Age ratio looks like it might be a risk factor.

df["Age_risk"] = (df["Age"] < 34).astype(int)
df["HourlyRate_risk"] = (df["HourlyRate"] < 60).astype(int)
df["Distance_risk"] = (df["DistanceFromHome"] >= 20).astype(int)
df["YearsAtCo_risk"] = (df["YearsAtCompany"] < 4).astype(int)

df["AttritionRisk"] = df["Age_risk"] + df["HourlyRate_risk"] + df["Distance_risk"] + df["YearsAtCo_risk"]

最终决胜代码是在一个baseline上添加了这些：
kaggle discussion - 1st place. That was unexpected…

df['MonthlyIncome/Age'] = df['MonthlyIncome'] / df['Age']

df["Age_risk"] = (df["Age"] < 34).astype(int)
df["HourlyRate_risk"] = (df["HourlyRate"] < 60).astype(int)
df["Distance_risk"] = (df["DistanceFromHome"] >= 20).astype(int)
df["YearsAtCo_risk"] = (df["YearsAtCompany"] < 4).astype(int)

df['NumCompaniesWorked'] = df['NumCompaniesWorked'].replace(0, 1)
df['AverageTenure'] = df["TotalWorkingYears"] / df["NumCompaniesWorked"]
# df['YearsAboveAvgTenure'] = df['YearsAtCompany'] - df['AverageTenure']

df['JobHopper'] = ((df["NumCompaniesWorked"] > 2) & (df["AverageTenure"] < 2.0)).astype(int)

df["AttritionRisk"] = df["Age_risk"] + df["HourlyRate_risk"] + df["Distance_risk"] + df["YearsAtCo_risk"] + df['JobHopper']