Forecasting Voting Trend of Election Campaign Using Hybrid Machine Learning Algorithms

Résumé

This paper proposes and evaluates a hybrid machine-learning framework to forecast short
term voting tendencies during election campaigns. The framework combines a genetic-algorithm (GA) based
feature-selection stage with a soft-voting ensemble classifier comprising logistic regression, random forest, and
gradient boosting models. The method is demonstrated on a synthetic but realistic electoral dataset that
incorporates polling averages, social-media sentiment proxies, economic indicators, incumbency, turnout, and
regional dummy variables. The results show that the GA selects a compact and informative feature subset, and
the hybrid ensemble achieves strong predictive performance (example test accuracy ≈ 0.80–0.85; ROC AUC ≈
0.85–0.90). The study provides the complete code, dataset generation procedure, and saved outputs to support
reproducibility. The proposed approach is intended for exploratory forecasting and campaign monitoring, and
not as a replacement for rigorously designed probabilistic election models that incorporate complex sampling
and weighting procedures.