Key skills employees leave the company. This departure happens so suddenly that employee turnover becomes an unbearable and costly and destructive problem for the company. Why? Because recruiting and training an outside person who is completely unfamiliar with the details of the company requires a lot of time and money.

In the face of this situation, whenever your colleague leaves the office where you work, a problem always comes to your mind:

“What if we could predict who might leave and understand why?”

But these assumptions are somehow wrong before assuming that employee turnover is just a disconnect at work, or better learning/growth opportunities exist elsewhere.

So, no matter what happens in your office, you will see them going out more than they enter.

But if you don't observe a pattern, you're ignoring the full significance of the employee turnover that is happening in real time in your office.

You might be wondering: “Are the company and its HR departments trying to prevent valuable employees from leaving?”

Yes! Therefore, in this article, we will build a simple machine learning model to predict employee turnover and interpret the results using SHAP tools so that HR teams can act based on these insights.

Table of contents

• Problem understanding
• Analysis on how to help human resources respond proactively
• What is the SHAP method?
• Dataset Overview
• 5 steps to SHAP tools/methods
• Revise the plan
• in conclusion
• FAQ

Problem understanding

In 2024, the market data report released by WorldMetrics clearly stated that 33% of employees left because they did not see opportunities for career development—that is, one-third of them left because of stagnant growth paths. Therefore, out of 180 employees each year, 60 employees will resign and leave the company. So, what is employee turnover? You might ask us this.

• What is employee turnover?

Gartner, a world-renowned consulting firm, has provided insight and professional guidance to clients over the past 45 years, defining employee turnover as “the gradual loss of employees when positions are not filled, often due to voluntary resignation, retirement or internal transfer.”

Analysis on how to help human resources respond proactively

The HR department is extremely reliable and valuable to the company because HR is the only department that can actively handle employee turnover analysis and HR.

HR departments can use analysis to identify the root causes of employee turnover, identify patterns/demographics of historical employee data models, and design targeted actions based on this.

So, what methods/pathways are helpful to the HR department? Guess it? The answer is the SHAP method. So what exactly is it?

What is the SHAP method?

SHAP is a method and tool for explaining machine learning (ML) model output.

It can also explain what causes employees to voluntarily resign, which you will see in the article below.

But before that, you can install it through the pip terminal and the conda terminal.

 <code>!pip install shap</code>

or

 <code>conda install -c conda-forge shap</code>

IBM presented a dataset called "IBM HR Analytics Employee Attrition & Performance" in 2017, using SHAP tools/methods.

So, the following is a brief overview of the dataset, you can refer to the following content,

Dataset Overview

We will use the IBM HR Analytics Employee Attrition dataset. The dataset contains information about more than 1,400 employees—including age, salary, position roles, and satisfaction scores—to identify patterns using SHAP methods/tools.

We will then use the following key columns:

• Attrition : Whether the employee leaves
• Overtime, job satisfaction, monthly income, work and life balance

After that, you should actually apply the SHAP method/tool through the following 5 steps to overcome the risk of employee churn.

5 steps to SHAP tools/methods

Step 1: Load and explore the data

 <code>import pandas as pd</code> from sklearn.model_selection import train_test_split<p> from sklearn.preprocessing import LabelEncoder</p><h1> Loading the dataset</h1><p> df = pd.read_csv('WA <em>Fn-UseC</em> -HR-Employee-Attrition.csv')</p><h1> Basic Exploration</h1><p> print("Shape of dataset:", df.shape)</p><p> print("Resignation value statistics:\n", df['Attrition'].value_counts())</p>

Step 2: Preprocess the data

Once the dataset is loaded, we convert the text value to numbers and divide the data into training and testing sections.

 <code># 將目標變量轉換為二進制</code><p>df['Attrition'] = df['Attrition'].map({'Yes': 1, 'No': 0})</p><h1> Encode all classification characteristics</h1><p> label_enc = LabelEncoder()</p><p> category_cols = df.select_dtypes(include=['object']).columns</p><p> for col in category_cols:</p><p> df[col] = label_enc.fit_transform(df[col])</p><h1> Define features and goals</h1><p> X = df.drop('Attrition', axis=1)</p><p> y = df['Attrition']</p><h1> Divide the dataset into training and testing</h1><p> X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)</p>

Step 3: Build the model

Now, we will use XGBoost, a fast and accurate machine learning model for evaluation.

 <code>from xgboost import XGBClassifier</code><p> from sklearn.metrics import classification_report</p><h1> Initialize and train the model</h1><p> model = XGBClassifier(use_label_encoder=False, eval_metric='logloss')</p><p> model.fit(X_train, y_train)</p><h1> Predict and evaluate</h1><p> y_pred = model.predict(X_test)</p><p> print("Classification Report:\n", classification_report(y_test, y_pred))</p>

Step 4: Use SHAP to interpret the model

SHAP (SHapley Additive exPlanations) helps us understand which characteristics/factors are most important in predicting churn rates.

 <code>import shap</code><h1> Initialize SHAP</h1><p> shap.initjs()</p><h1> Explain model predictions</h1><p> explainer = shap.Explainer(model)</p><p> shap_values = explainer(X_test)</p><h1> Summary</h1><p> shap.summary_plot(shap_values, X_test)</p>

Step 5: Visualize Key Relationships

We will use SHAP dependency graphs or seaborn to visualize resignation and overtime.

 <code>import seaborn as sns</code><p> import matplotlib.pyplot as plt</p><h1> Visualize resignation and overtime</h1><p> plt.figure(figsize=(8, 5))</p><p> sns.countplot(x='OverTime', hue='Attrition', data=df)</p><p> plt.title("Resignation and overtime")</p><p> plt.xlabel("Overtime Situation")</p><p> plt.ylabel("Count")</p><p> plt.show()</p>

Output:

Now, let's shift our focus to 5 business insights from our data

However, of the 5 insights, 3 key insights from IBM datasets based on SHAP approaches are something that companies and HR departments should actively pay attention to.

3 key insights from IBM datasets based on SHAP method:

1. Employees who work overtime are more likely to leave.
2. Low job and environmental satisfaction increases the risk of churn.
3. Monthly income also has an impact, but it is not as significant as overtime and job satisfaction.

Therefore, the HR department can use the above insights to find better solutions.

Revise the plan

Now that we know what factors are important, the HR department can follow these 4 solutions to guide HR policies.

1. Revisiting compensation plans

Employees have families to support, bills to pay, and lifestyle to maintain. If companies don’t revisit their compensation plans, they are likely to lose their employees and face competitive disadvantages in their business.

2. Reduce overtime or provide incentives

Sometimes, work can be waited, but stress cannot. Why? Because overtime does not mean motivation. A nervous shoulder without motivation can lead to various insecurity and health problems.

3. Improve job satisfaction through employee feedback

Feedback is not just a tool to move forward, it is also an implementation cycle/guideline for what should be an invisible future. If employee turnover is a problem, then employees are the solution. Asking helps, assuming it will erode.

4. Promote a better balance between work and life

People join work not only because of social pressure, but also to discover their true selves and their abilities. Finding a job that meets both goals can help improve their productivity; however, over-utilizing skills can be counterproductive and illogical for companies.

Therefore, SHAP-based method datasets are very suitable:

• Churn Prediction
• Labor Optimization
• Interpretable AI Tutorial (SHAP/LIME)
• Feature importance visualization
• Human Resources Analysis Dashboard

in conclusion

Predicting employee turnover can help companies retain the best employees and thus maximize profits. Therefore, through machine learning and SHAP, the company

The above is the detailed content of Predict Employee Attrition with SHAP: An HR Analytics Guide. For more information, please follow other related articles on the PHP Chinese website!