Employee attrition prediction for imbalanced data using genetic algorithm-based parameter optimization of XGB Classifier

Attrition of employees is vital for any organization as it significantly influences productivity and hampers the long-term growth strategies of the organization. Since employee attrition leads to loss of skills and experiences any organization always try to find a way to retain their employees to reduce training and recruiting cost as well as to achieve their business goal smoothly. Machine learning approaches, which predict the possibility of attrition based on the employee attributes avoid the tedious, and biased manual prediction, and help the organization take preventive measures. This paper presents a framework for attrition prediction that emphasizes imbalance classification and the adoption of genetic algorithms to optimize the model. First, we have adopted different oversampling methods like Synthetic Minority Over-sampling Technique (SMOTE), Adaptive Synthetic (ADASYN), and Borderline Synthetic Minority Over-sampling Technique to balance our data set. We have used XGBoost classifiers for classification with the data that are obtained from different over-sampling techniques. As the XGBoost classifier has many hyperparameter a genetic algorithm is used to optimize our model where the accuracy is chosen as the fitness function. The comparative performance analysis of different over-sampling methods as well as hyper-parameter tuning (Amongst Genetic algorithm, GridSearchCV, and with the default value of different hyper-parameter) on the real dataset suggests that SMOTE for oversampling techniques and genetic algorithm for optimization attains improved performance.