This project builds a neural network model to predict employee attrition and the most suitable department for each employee in a company using TensorFlow and Keras. The model uses a branched architecture to simultaneously predict binary attrition status and multi-class department affiliation.
The dataset used (attrition.csv) includes various employee characteristics such as age, job satisfaction, education, and more. The dataset consists of 1470 samples with 27 features.
The preprocessing steps include:
- Encoding categorical variables ('Attrition', 'OverTime', 'Department') using
OneHotEncoder. - Scaling numerical features with
StandardScalerto normalize the data. - Splitting data into training and testing sets.
The model features:
- An input layer that takes in the scaled features.
- Two shared dense layers.
- Two branches:
- Attrition Branch: Predicts employee attrition with a softmax output layer.
- Department Branch: Predicts department classification with a softmax output layer.
- The model is compiled with the
adamoptimizer andcategorical_crossentropyloss for both outputs. - It is trained for 50 epochs with a batch size of 32.
- Model performance is evaluated on the test set with accuracy as the metric for both outputs.
To run this model, you will need Python 3 and the following libraries:
pandasnumpytensorflowsklearn
- Attrition prediction accuracy: 81.79%
- Department classification accuracy: 51.90%
- Implementing cross-validation to assess model stability.
- Adding dropout layers to prevent overfitting.
- Tuning hyperparameters like learning rate and number of neurons.
This neural network model serves as a robust tool for predicting employee attrition and department suitability, aiding HR decisions in corporate environments.