Path-Loss Prediction Model for Wireless Communication in Eagle Island Using Feed-Forward Artificial Neural Network

This research investigated the development and application of a path-loss prediction model for wireless communication networks using an Artificial Neural Network (ANN), with a focus on the specific geographical and environmental context of Eagle Island, Port Harcourt. The study is motivated by the need for more accurate path-loss predictions in diverse and dynamic settings, which traditional empirical models, such as the Okumura model, may not fully address. To achieve this, the ANN model was trained on a rich dataset that incorporated various weather conditions and frequencies ranging from 1 GHz to 4 GHz, capturing a broad spectrum of potential real-world scenarios. The training dataset was designed to reflect the unique environmental characteristics of Eagle Island, including variations in terrain, vegetation, and atmospheric conditions. This comprehensive approach allowed the ANN to learn intricate patterns and relationships within the data, enhancing its predictive accuracy. The performance of the ANN model was rigorously compared with the Okumura model, which serves as a benchmark due to its widespread use in path loss prediction. The comparative analysis revealed that the ANN model provided superior path loss predictions relative to the Okumura model, achieving a Root Mean Square Error (RMSE) of 0.00080506, a Mean Absolute Error (MAE) of 0.00061085, a correlation coefficient (R) of 0.99999, and a coefficient of determination (R²) of 0.99999. In contrast, the Okumura model recorded significantly higher errors (RMSE = 0.46151, MAE = 0.40561) and a negative R² value (–1.6114), indicating poor model fit. The ANN’s adaptability across different frequencies and weather conditions enabled it to deliver highly precise and reliable predictions, effectively capturing the complex, non-linear relationships that the Okumura model failed to represent. This superior performance highlights the ANN’s strong potential for improving network planning and optimization through a more adaptive and data-driven approach to path loss estimation. Overall, the findings demonstrate the value of integrating advanced machine learning techniques into wireless communication modeling. The ANN’s proven accuracy and generalization capacity make it a dependable tool for designing and managing wireless networks in diverse and challenging environments such as Eagle Island, while also laying the groundwork for further applications of machine learning in radio propagation analysis.