Abstract

A study on the Levenberg-Marquardt Neural Network model was aimed at predicting UHF band services spectrum occupancy in the Osogbo metropolis, Nigeria. Four locations were used: Firestation, Uniosun campus, National Control Centre (NCC), and Fountain University campus. Energy detection was used for data capture and analysis. The prediction metrics included Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE). Parametric values considered included channel vacancy duration versus time, Power Spectral density versus time, Channel vacancy duration versus frequency, Power spectral density versus frequency, power spectral density versus channel vacancy duration (Time Domain), and power spectral density versus channel vacancy duration (Frequency Domain). The highest RMSE and MAPE recorded were 39.94 and 28.59 %, respectively, for Fountain University, while the lowest RMSE and MAPE were 0.269 and 0.09 %, respectively, for UNIOSUN. The fire station and National Control Centre locations had 0 RMSE and 0% MAPE, which were the most accurate predictions of all parametric values.