END TO END CAMPUS PLACEMENT PREDICTION SYSTEM USING ENSEMBLE LEARNING

Abstract

Campus placement is one of the most important outcomes of an undergraduate engineering programme, both for the students who hope to begin their careers and for the institution whose reputation depends on its placement record. Training and placement cells spend a great deal of effort preparing students for recruitment drives, yet they usually learn which students are at risk of remaining unplaced only after the recruitment season has ended. This paper presents an end to end campus placement prediction system that uses ensemble learning to estimate, well before the drives begin, how likely each student is to be placed. The system is built on academic and profile data that institutions already hold. The attributes considered include secondary and higher secondary marks, semester grade point averages, the number of active and cleared backlogs, branch of study, internship experience, certifications, participation in technical events, aptitude test scores, communication assessment scores, and the results of mock interviews. These records are collected from the examination section and the placement cell, cleaned, encoded, and combined into a single structured dataset in which every row describes one student. Several individual classifiers are trained on this dataset, including logistic regression, a decision tree, k-nearest neighbours, and a support vector machine, so that a baseline can be established. Ensemble methods are then applied, namely a random forest, gradient boosting, an extreme gradient boosting model, and a soft-voting ensemble that combines the strongest individual learners. The models are compared using accuracy, precision, recall, F1 score, and the area under the ROC curve under stratified cross-validation, and the ensemble that gives the most balanced performance is selected for deployment. In addition to a yes or no prediction, the system estimates the expected salary band for students who are predicted to be placed and reports the factors that most influence each prediction. Feature importance and per-student explanations show, for example, whether a student's weak point lies in aptitude, in communication, or in the lack of practical exposure. This information is presented through a web dashboard in which the placement officer can view the overall readiness of a batch, filter students by department, and open an individual profile to see the recommended areas of improvement. The complete pipeline, from data ingestion to the dashboard, is implemented in Python with commonly used open-source libraries, so that it can be maintained by the institution without specialised infrastructure. The objective is not to label students but to help the placement cell direct its training resources towards those who need them most, early enough for the intervention to make a difference. Future extensions may include company-specific eligibility matching, integration with learning management systems, and periodic retraining as new placement seasons add fresh data.