Barhrhouj, ayah (University Aix Marseille III), Ananou, Bouchra (LSIS), Ouladsine, Mustapha (Université d'aix marseille III)

Balancing Feature Selection, Model Accuracy, and Transparency in Maritime Machine Learning: A Trade-Off Analysis

Scheduled for presentation during the Regular Session "Modelling and simulation" (WeCB), Wednesday, June 11, 2025, 16:50−17:10,

33rd Mediterranean Conference on Control and Automation, June 10-13, 2025, Tangier, Morocco

This information is tentative and subject to change. Compiled on May 9, 2025

Abstract

Feature selection (FS) is a critical step in ma- chine learning (ML) applications, particularly in the maritime transportation domain, where large-scale data from sensors, weather conditions, and operational logs can introduce re- dundancy and noise. Feature selection plays a crucial role in improving ML model performance by reducing dimen- sionality, enhancing generalization, and mitigating overfitting. However, its impact extends beyond predictive accuracy to the explainability of AI (XAI), influencing how interpretable and transparent models become. In this study, we systematically compare various feature selection methods—including filter, wrapper, and embedded techniques—analyzing their effects on model performance and explainability. We assess multiple ML models across different datasets to evaluate trade-offs between predictive accuracy, computational efficiency, and ex- plainability. The study leverages shapley additive explanations to compute feature importance scores to quantify how feature selection impacts model transparency. Our findings highlight that while some FS methods enhance predictive accuracy, they may compromise interpretability, whereas others strike a balance between accuracy and explainability.