Analysis of the Effectiveness of Neural Networks with Different Configurations

In the dynamic landscape of contemporary organisations, employee retention emerges as a pivotal factor for stability and successful development. This study focuses on the application of neural networks to analyse factors influencing employees’ decisions to stay or leave a company. Utilising an anonymized dataset encompassing education, work experience, demographics, and employment history, various neural network configurations were explored, adjusting the number of layers and neurons while maintaining consistent parameters like sigmoid activation function and Resilient Propagation algorithm. The research revealed a noteworthy trend: an increase in the number of neurons enhances model performance, yet saturation occurs beyond a certain point. The model with a single hidden layer and 64 neurons exhibited optimal results, striking a balance between model complexity and prevention of overfitting. Beyond model error, crucial steps like cross-validation and optimal hyperparameter tuning are discussed to ensure a robust analysis. The study concludes by providing a foundation for future work in employee attrition prediction, emphasising the significance of a unified approach for model comparison.