Sonar Signal Classification through Dimensionality Reduction and Artificial Intelligence: A Comparative Study

This study examines the impact of dimensionality reduction techniques, particularly Principal Component Analysis (PCA) and Sparse Principal Component Analysis (SPCA), on the performance of various classification models. Utilizing a dataset comprised of sonar signals intended to distinguish between rocks and mines, we explore the efficacy of these techniques across multiple models including Random Forest (RF), Support Vector Machine (SVM), XGBoost, and a Feed-forward Neural Network (NN). Our experiments are structured into three distinct scenarios: direct application of classifiers, application following PCA, and application following SPCA. The findings reveal that SVM and RF excel without dimensionality reduction, achieving high accuracy rates. However, the application of SPCA notably enhances the performance of the NN, marking a significant improvement in metrics and efficiency, while SVM and RF experience a decrease in effectiveness with dimensionality reduction. Conversely, XGBoost shows consistent performance across all scenarios. This research underscores the variable influence of dimensionality reduction on different classifiers and suggests that the choice of technique should be tailored to the specific characteristics of both the model and the data.