Predictive Deep Learning approach of employee attrition for imbalance datasets using SVMSMOTE algorithm with Bias Initializer

Abstract

Abstract Nowadays the biggest problem with all the corporate firms is to retain their employees. After the pandemic this problem arose with very high attrition. The statistic says the employee attrition worldwide is approximately 15 to 20% and in India average 22%. Employee attrition is still a very severe problem and all the organization must be aware of it. Therefore, to retain the good and reliable employees of the organization, and to continue growth in the competitive business environment organization should minimize the attrition of employees. Systems need to maintain their employee for they have to require a system which is based on some prediction. These predictive models’ give the result why the right people leave the organisation and what are the major reasons behind attrition. In every organization that is under the human resource department, they have to deal with the employee and they must know the reason or criteria behind the attrition. This prediction based technique gives support to them to find the attributes like salary, promotion, position and satisfaction are lacking and improve to retain the best and experienced employees. This model uses the prediction of employees based on whether the dataset is balanced or not. The proposed algorithm uses deep learning with the steps of data acquisition, exploration, pre-processing and feature selection on training models. A deep learning approach is suggested in this study as a way to deal with class imbalance through resampling minority classes using SVM SMOTE (Support Vector Machine Synthetic Minority Oversampling Technique) by adding a Bias initializer to the output Layer. And with this the accuracy, recall, precision and f1 Score output are respectively 89%, 92%, 94% and 93% achieved.