PERANCANGAN MODEL PREDIKSI CHURN KARYAWAN DI PT X MENGGUNAKAN TEKNIK DATA MINING

• Main Author: Rizmadanty Singgih, Tsaqila
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/51194
• Institution: Institut Teknologi Bandung

PT X is a financial technology developer company in Indonesia. PT X faces a relatively high employee attrition rate which has reached fifteen percent. PT X has fifteen percent attrition rate. The rate is much higher than the average attrition rate in Indonesia which is seven percent (Mercer, 2020). High attrition rate has been a problem for many companies. High attrition rate in Pt X causes several internal problems for the company. PT X must pay a huge cost for employee recruitment processes and trainings for new hires. PT X also must run its business with many vacant positions in the company during the turnover process. Moreover, it takes time for new hires to fully adapt with new environments and to have a level amount of knowledge with their predecessors. Setting off from these problems, this research aimed to build an employee churn prediction model and to develop a simple application to run the prediction model. The research is conducted through a series of steps developed based on Cross Industry Standard Process for Data Mining (CRISP-DM). The prediction model is built under one of three alternative algorithms, which are decision tree, random forest, and gradient boosting. The best prediction model is selected based on five performance criteria, which are accuracy, precision, recall, F1, and ROC scores. From three alternative models, prediction model built under random forest algorithm with 37 predictors is selected as the best model. This model achieves relatively high-performance scores in all performance criteria. The model scores 0.904 for accuracy; 0. 894 for precision; 0.908 for recall; 0.901 for F1; and 0.904 for ROC. The selected prediction model is integrated into a simple application using Microsoft Excel. The application helps users to execute the prediction model and see the prediction result.