Utilizing a Hybrid Deep Learning Framework Model with Bias-Resistant Inputs on Improved Election Prediction Accuracy

Elections have become more difficult to model than ever due to the rise of social media and misinformation. Predictions remain essential as they inform future policy decisions. This study proposes a hybrid approach designed to improve election prediction accuracy by integrating a bias-aware and bias-resistant architecture with a deep learning framework. This model incorporates economic and demographic trends along with 11 binary factors. It was hypothesized that this approach would give more accurate predictions than traditional methods. Data were sourced from Yahoo Finance, the US Census Bureau, Kaggle, and Newberry (2024). The model was trained and tested on historical data, cross-validated, and ran on 100 simulations. Predictions were compared to the 2024 US congressional election results. A one-sample t-test under a 95% confidence interval indicated that every factor was significant, thereby supporting that the data and hybrid architecture meaningfully contributed to the model’s accuracy and efficiency.