SPECTRUM OCCUPANCY PREDICTION USING A MODIFIED XGBOOST MACHINE LEARNING ALGORITHM

Abstract

There is an alarming rate of growth in the usage of spectrum, where some of the allocated spectra is fully engaged while others are sparsely utilized. This gives attention to the use of cognitive radios where the Primary users can maximize the available spectrum holes alongside the secondary users. The challenge of using cognitive radio technology is interference which is a factor that causes a delay in the handoff time. XGBoost alongside other regression Machine Learning (ML) Algorithm such as linear Regression, Lasso Regression, Ridge regression, and the random forest was used to train and predict the dataset gotten from sensing the spectrum at a location called Morris Fertilizer within the environs of Minna, Niger state. Linear regression, Random forest regression, XGBoost, Ridge, and Lasso have been used for the prediction of cognitive radio frequencies based on 10 power features. The linear Regression, Ridge and Lasso gave the same level of accuracy of 6.39%, while Random forest gave an accuracy of 54.65%Xgboost gave the best performance with an accuracy level of 96.85%, thus boosting algorithm shows a high level of prediction ability.