Predicting Voting Patterns Using Generative Adversarial Networks and Stack Star Classifier

This study explores the impact of parental political involvement and affiliations on the voting behaviour of young adults in Ghana. The research had two key objectives: first, to develop a comprehensive dataset within the Ghanaian context, incorporating demographic and socioeconomic factors that influence parental political participation, and second, to leverage this dataset to build a novel predictive model, Stack Star (Stack*), designed to forecast young adults' voting patterns based on their parent's political affiliations. The Stack* model integrates a range of machine learning classifiers, including Random Forest, Naive Bayes, K-Nearest Neighbour, and Support Vector Machine as base classifiers, with Logistic Regression serving as the meta-classifier. A grid search approach was employed for hyperparameter tuning to optimize the performance of the classifiers. The dataset, comprising parents' political activities, was collected through Google-guided surveys. Extensive experimental results demonstrated that the Stack* model outperforms state-of-the-art single classifiers across all evaluation metrics, achieving a remarkable accuracy of 91.75% and a mean absolute error of 0.22. These findings underscore the significance of parental political behaviour as a predictor of young adult voting patterns and highlight the efficacy of machine learning techniques in political behaviour forecasting.