Candidate Joining Prediction Using Random Forest and KNN

This paper presents a web-based machine-learning system for predicting whether a selected candidate is likely to join an organization after accepting an employment offer. The study uses 8,995 historical recruitment records with fifteen candidate- and offer-related factors, including notice period, offer-acceptance duration, compensation changes, relocation, candidate source, experience, location, and age. Categorical attributes are encoded and the dataset is divided using a stratified 80:20 train-test split. Random Forest and K-Nearest Neighbors (KNN) classifiers are trained and compared. The Random Forest model, configured with 250 trees and balanced subsampling, achieves 82.27% accuracy with a recorded training time of 3.12 seconds, while the optimized KNN model achieves 80.71% accuracy with a training time of 15.21 seconds. The selected Random Forest model is integrated with a Flask, MySQL, Jinja, Bootstrap, and pandas web application that supports administrator and HR-user workflows, dataset viewing, prediction, model comparison, graphs, and query management. Although the system demonstrates successful end-to-end integration, the confusion matrix shows low recall for the Not Joined class. Therefore, the output is intended as decision support for recruiter follow-up rather than an automated employment decision