ENHANCING ECONOMIC FORECASTING WITH BAYESIAN NEURAL NETWORKS: A FOCUS ON GDP PREDICTION AND UNCERTAINTY QUANTIFICATION IN NIGERIA

This study produces a Bayesian Neural Network (BNN) to forecast GDP growth while quantifying prediction uncertainty which is often overlooked in traditional neural networks. By integrating Bayesian inference with deep learning, the model generates prediction intervals rather than single-point estimates, offering a probabilistic perspective crucial for decision-making under uncertainty. The BNN was initially trained on simulated data to validate its architecture and subsequently tested on real-world quarterly GDP data (2010-2023), monthly inflation rates, and interbank interest rates sourced from the Central Bank of Nigeria. The model employs Guassiandistributed weights and biases, uses Rectified Linear Unit (ReLU) activation functions, and optimizes training through the Adam algorithm. Results demonstrates that the BNN achieves strong predictive performance, with its prediction interval providing actionable insights for scenarios where uncertainty quantification is paramount. This approach improves GDP forecasting accuracy and provides a robust framework for analyzing volatile economic metrics in emerging economies like Nigeria.