Intelligent Analysis of TV Spectrum Occupancy in Cognitive Radio Networks Using Machine Learning

Spectrum sensing plays a crucial role in enabling unlicensed secondary users to utilize unoccupied spectrum within cognitive radio networks. This research concentrated on the methods by which new secondary users can identify available spectrum opportunities. In this investigation, we collected data from ten different locations surrounding Makerere University, resulting in a dataset that illustrates spectrum utilization, particularly emphasizing the varying power levels across different frequencies. We analyze the utilization of television spectrum within cognitive radio networks by utilizing supervised machine learning techniques, particularly decision trees, support vector machines, and linear regression. The research involved the manual collection of one thousand data sets from each location, resulting in a cumulative total of 10,000 samples. The machine learning algorithms employed included Logistic Regression, K-Nearest Neighbors, and Naive Bayes. The application packages utilized were Pandas, NumPy, Scikit-learn, and Matplotlib, which facilitated the loading of the file, data visualization, data manipulation, and the generation of high-quality graphs, charts, and plots. The dataset was divided into two parts: the training set and the testing set, with 80% designated for training and 20% for testing. The Logistic Regression algorithm exhibited the highest performance, achieving an accuracy and precision of 0.9904, with its learning curve revealing a strong correlation between the training set and the test set. The K-Nearest Neighbor algorithm ranked second in performance, also attaining an accuracy and precision of 0.9904, with its learning curve revealing a fairly close relationship between the training set and the test set. The Naive Bayes algorithm demonstrated the lowest performance, with an accuracy and precision of 0.95, and its learning curve illustrated a relatively poor relationship between the training set and the test set when compared to the other two algorithms. Conclusively, logistic Regression achieved the highest level of accuracy, making it the most efficient model for this study.