SIGNAL CLASSIFICATION FOR JAMMING DETECTION: A COMPARATIVE STUDY OF MACHINE LEARNING AND DEEP LEARNING APPROACHES

Abstract

This study investigates the effectiveness of several machine learning and deep learning algorithms in a binary classification between jamming and non-jamming signals. Models under scrutiny include Support Vector Machine, Logistic Regression, Random Forest, K-Nearest Neighbors, Deep Neural Network and Long Short-Term Memory networks. The results indicate that traditional machine learning models, specifically Support Vector Machine, performed very well as compared to deep learning models, which resulted in an accuracy of 96.97% and AUC of 0.99. Results for Random Forest and K-Nearest Neighbors are also promising with high precision and favorable recall. In contrast, deep learning models, such as Deep Neural Network and Long Short-Term Memory, performed badly with small datasets. Long Short-Term Memory had a good recall of 99.73%, poor precision at 48.86%, whereas Deep Neural Network performed at low accuracy of 47.90% and AUC of 0.47. These outcomes clearly point out the fact that the machine learning models were highly adaptable in small dataset scenarios and therefore could be used more effectively for resource-constrained classification problems. Although deep learning models have great promise, they require large amounts of data and advanced preprocessing for peak performance. This study has underlined the continued relevance and effectiveness of machine learning models in the signal classification domain for practical applications.