ML Based Event Analytics and Participation Prediction System

Predicting whether a registered participant will actually attend academic events is a common problem that most institutions leave unsolved—not because the data is unavailable,but because no lightweight, interpretable tool exists to analyseevent structures. This paper presents EventIQ v2, a machine learning system designed specifically for the small, heterogeneousdatasets typical of institutional settings. The system introduces a seven-term interaction feature layer—including rating xorg,buzz xrating, and loyalty score—that encodes how event quality,organiser reputation, and social engagement combine to drive attendance, rather than treating each factor in isolation. A Gradient Boosting Classifier trained on 3,000 records with 18 engineeredfeatures achieves 88% accuracy, an AUC-ROC of 0.956, and an F1 Score of 0.904. Five-fold cross-validation confirms stable generalisation (CV accuracy ≈88%), closing a 25-point gap from the v1 prototype. An ablation study shows the interaction layer alone contributes an AUC-ROC gain of 0.030. The system is deployed as a Flask REST API supplemented by a five-page Streamlit interactive dashboard with per-prediction explanation charts, making model output legible to faculty coordinators and student organisers without machine learning background.