Study and Prediction Analysis of the Employee Turnover using Machine Learning Approaches

Employee Turnover is one of the key market challenges in Human Resource (HR) Analytics. Organizations usually invest a greater amount of money and time in the hiring of staff and nursing them in the hope to receive value addition. When an employee leaves the company, the reduction of opportunity costs is borne by the company. Turnover is especially prevalent in large-scale recruitment agencies. The risk of replacing workers remains important for most employers. This is due to the amount of time spent recruiting and selecting a successor, the sign-on incentives, and the lack of morale for several months as the new employee gets used to the new job. The tangible costs of workforce turnover will be the cost of recruiting new staff, the cost of recruitment and hiring, the time of transition, future product or service quality issues, the cost of temporary staff, the cost of training, the cost of lack of production, the cost of lost expertise and the cost of the job being empty before an acceptable replacement is found. We find that the attributes of workers such as Job Position, overtime, work level affect significantly attrition. Various classification methods are introduced such as logistic regression, linear discriminate analysis, ridge classification, lasso classification, decision trees, random forests to forecast and concurrently measure the likelihood of turnover of every new employee. Data from an HR department of the company available at Kaggle were used to estimate the employee turnover. The dataset includes 10 different attributes of 1470 personnel. Dataset specifies if the personnel is leaving or staying based on the attributes. Now, to construct a prediction model based on the previously mentioned machine learning algorithms with 90 percent of the total personnel's attributes and the rest for model testing. The best performing performance algorithm yields the best accuracy of Random Forest is 90.20 percent and the worst accuracy of Naïve Bayes is 80.20 percentage.