Lasso Regularization Based Mutual Information Feature Selection for Acoustic Seabed Sediment Classification

Accurate classification of seabed sediments is critical for supporting various applications in marine research and resource exploration. However, the high dimensionality and complexity of acoustic data used for seabed mapping present challenges in selecting the most relevant and non-redundant features for classification tasks. To address this, we develop an approach that applies mutual information (MI) to evaluate and rank features based on their relevance to the target variable. Subsequently, Lasso regularization is introduced to refine the selected feature set by imposing sparsity, eliminating irrelevant and redundant features. This combined method ensures that the selected features are both informative and efficient, enhancing model generalization and reducing the risk of overfitting. Through comprehensive experiments on real-world acoustic seabed data, the proposed method demonstrated obvious improvements in classification accuracy compared to conventional feature selection approaches. The integration of MI with Lasso not only optimized the feature selection process but also improved the overall performance of seabed sediment classification models. This approach offers a robust and efficient solution for handling large, high-dimensional acoustic datasets.