Abstract

Machine learning, a cornerstone of artificial intelligence, equips engineers and scientists with powerful tools for uncovering valuable insights from complex, large-scale datasets across diverse applications. Engineering applications increasingly demand transparency and interpretability, particularly to understand the reasoning behind specific outcomes and decisions. This chapter addresses this need by exploring dimensionality reduction techniques that enhance model interpretability by focusing on the most significant variables. By employing robust, outlier-resistant methods, we enable a closer analysis of how individual data points impact results, making these approaches ideal for real-world data with inherent noise or contamination. The utility of these techniques is demonstrated on two datasets, one from gene expression analysis and the other from credit scoring. Furthermore, we discuss the relationship between dimensionality reduction and energy-efficient approximate computation, highlighting its relevance for engineering applications in AI.