Bayesian Model for Predicting Voting Behavior in Presidential Elections in Kenya
Onsongo Wycliffe Nyaundi *
Department of Research and Development, Delmonte Kenya Limited, Kenya.
Wycliffe Cheruiyot
Faculty of Science and Technology, Multimedia University, Kenya.
Antony Karanjah
Faculty of Science and Technology, Multimedia University, Kenya.
*Author to whom correspondence should be addressed.
Abstract
This study applies the Bayesian Dirichlet-Multinomial model to predict Kenya’s 2022 presidential election. The model incorporates voter age, gender, poverty rate, and political ideation as major determinants of voting behavior. The model estimated Raila Odinga’s vote share at 48.17% (actual: 48.85%) and William Ruto’s at 46.96% (actual: 50.49%). It slightly overestimated minor candidates but proved more accurate than polls. A Bayesian p-value of 0.487 confirmed model reliability. Bayesian inference demonstrated superior adaptability by quantifying uncertainty and updating probabilities. Key voter behavior determinants included youthful population and political ideation. The study concludes that Bayesian modeling enhances election forecasting and recommends integrating it with machine learning, social media sentiment analysis, and economic indicators for improved predictions.
Keywords: Presidential elections, election forecasting, bayesian prediction models