The increasing demand for wireless communication services has intensified concerns regarding spectrum scarcity and inefficient utilisation of available radio-frequency resources. Cognitive Radio Networks (CRNs) provide a solution through dynamic spectrum access, with spectrum sensing serving as a critical process for identifying vacant and occupied frequency bands. However, conventional sensing techniques can experience reduced detection performance and increased false alarm rates, particularly under low signal-to-noise ratio (SNR), noise uncertainty, fading, and dynamic channel conditions. This review examines the application of machine learning (ML) and deep learning (DL) techniques for improving spectrum sensing accuracy and reducing false alarm rates in cognitive radio networks. A comparative review of recent literature was conducted, focusing on studies published between 2023 and 2026. Relevant studies were analysed according to the learning technique employed, sensing environment, datasets, SNR conditions, probability of detection, probability of false alarm, accuracy, F1-score, and other reported performance measures. The reviewed evidence shows that both conventional ML and DL approaches can provide substantial improvements over traditional sensing methods, although performance varies according to experimental conditions and model architecture. In a comparative study, Random Forest achieved 97.17% accuracy, 95.7% probability of detection, and 96.93% F1-score, while demonstrating the lowest false alarm and missed-detection probabilities among the evaluated models. A U-Net-based approach reported a 99% detection probability and 5% false alarm rate at 10 dB SNR, while LSTM-based approaches demonstrated improved robustness under low-SNR conditions. Overall, the review establishes that ML-based spectrum sensing offers considerable potential for improving sensing reliability and spectrum utilisation, with adaptive, deep-learning, collaborative, and reinforcement-learning approaches emerging as promising directions for intelligent cognitive radio networks.

Keywords: Machine Learning, Spectrum Sensing Accuracy, Reduced False Alarm Rate and Cognitive Radio Networks.