Abstract—Cognitive Radio (CR) technology enhances spectrum efficiency by enabling unlicensed users to utilize underused frequency bands opportunistically. This work introduces a methodology that integrates deep learningbased feature extraction with classification techniques grounded in machine learning to enhance decision-making within Cognitive Radio Networks (CRNs). Spectral characteristics are derived from image representations of power distribution and subsequently categorized using four classifiers: Binary Decision Tree (BDT), Discriminant Analysis Classifier (DAC), K-Nearest Neighbors (KNNC), and Support Vector Machines (SVM). The study’s main contribution lies in validating the applicability of deep neural networks for spectral classification and assessing the influence of each classification model on decision-making outcomes. Performance is measured through total and failed handoffs, average bandwidth, and overall delay. Findings indicate that the SVM classifier achieves superior accuracy and operational efficiency. The SVM classifier performed best, recording the fewest handoffs and the lowest average delay. Regarding the number of handoffs, BDT, KNNC, and DAC increased this metric by 1.44, 1.88, and 2.88 times, respectively, demonstrating the greater efficiency and stability of the SVM classifier. SVM also achieved the best results for the other metrics evaluated.