Employee Attrition Prediction

Employee attrition is a crucial aspect of human resource management, influencing strategic workforce planning and retention efforts. This comprehensive documentation guides you through the process of analyzing employee attrition using a dataset. The project encompasses data preprocessing, exploratory data analysis (EDA), and the implementation of various machine learning models.

V1 File

Project Steps

Dataset

The dataset contains employee information for predicting attrition, including attributes like age, daily rate, department, and more.

Data Cleaning

Columns with over 80% empty values were dropped to streamline the dataset.

Data Preprocessing

1. Encoding Categorical Columns: Categorical columns were encoded using LabelEncoder.
2. Normalizing Data: A Gaussian distribution was created for each attribute. Attributes with a standard deviation below 0.8 were excluded after normalization.

Exploratory Data Analysis (EDA)

Explored the dataset to derive insights into employee characteristics and potential factors contributing to attrition.

Principal Component Analysis (PCA)

Applied PCA for feature extraction and visualization.

Machine Learning Models

Implemented the following models:

• Mean-Shift: Utilizes Mean-Shift clustering for prediction. Achieves an accuracy of 86.17%.

• K-Nearest Neighbors (KNN): Employs a KNN classifier with k=3. Achieves an accuracy of 78.68%.

• Support Vector Machine (SVM): Implements an SVM classifier with a linear kernel. Achieves an accuracy of 86.17%.

• K-Means: Applies K-Means clustering for prediction. Achieves accuracy metrics, precision, recall, and F1 score.

• Decision Tree: Trains a Decision Tree classifier. Achieves an accuracy of 86.17%.

V2 File

Data Visualization and EDA

Explored through bar charts for feature comparison between different attributes to gain insights into employee attrition.

We have used Random Forest Classifier for best results of Bar charts

The Random Forest Classifier was employed to identify the top features contributing to employee attrition. Top 10 Features:

• 'OverTime'
• 'MonthlyIncome'
• 'TotalWorkingYears'
• 'Age'
• 'YearsAtCompany'
• 'StockOptionLevel'
• 'JobLevel'
• 'YearsWithCurrManager'
• 'MaritalStatus'
• 'JobRole'

Heatmap Analysis for Feature Selection

Encoding Categorical Labels

Categorical columns are encoded using LabelEncoder for numerical analysis.

Correlation Heatmap

A heatmap of the correlation matrix visualizes relationships between features, aiding in identifying potential correlations influencing employee attrition.

Filtered Heatmap for Top Features

Selected important features ('OverTime', 'MonthlyIncome', etc.) are analyzed in a filtered heatmap for deeper insights.