Predictive modeling of job offer acceptance for a major Indian IT company

Abstract

Talent recruitment is getting harder and harder because of the shrinking talent pool with relevant experience in most of the industries especially in IT. IT companies invest lot of time in doing talent search, spotting candidates with relevant experience, conducting tests, interviews, background check and health checkup, etc. before they offer the job to the potential employee. But from the job applicants perspective, they find a job and most of the time, they would have at least a few weeks between the offer date and the expected joining date, during which they shop around for still better job in terms of role, salary, location or they try to negotiate with the current employer to at least match the new offer that they have in hand. This results in the candidate either continuing with the current employer or some other employer who offered a better job than the one in hand. The real problem here is the employer who offered the job who would not even know until the expected date of joining about the whole situation. This puts the employer in tough situation and results in money loss, time loss, potential delay of crucial projects and customer dissatisfaction. This project made an attempt to build a model to predict whether the applicant is likely to join the company based on the factors such as qualification, skills, designation, total work experience, relevant work experience, joining location, number of days between offer made date and expected joining date, etc.. Initial data screening was done using histogram, box plots and graphical visualization to understand the distribution of values for all variables and to check if there is any direct correlation between the independent (explanatory) variables and dependent variable. Data was also checked for missing values and null and the data was prepared accordingly. Two different approaches have been taken for building the predictive model. First approach was to build the models using classification techniques like C5.0, Support Vector Machines (SVM), and Artificial Neural Networks(ANN) separately. Second approach was to build the hybrid models using combination of classification techniques like C5.0, SVM, ANN, Nearest Neighbor, Bayes Net Classifier and Classification and Regression Tree (CART). Some of the hybrid models performed better than the separate models. Models show that CTC Jump, number of days between offer made date and expected joining date and designation are more important factors compared to the others factors.