Identifying optimal technique of reducing dimensionality of scour influencing hydraulic parameters applying SWOT analysis

Abstract

Spur dike-like structures (SDLS) are built in rivers to control erosion and reroute water flow. Predicting scour depth around SDLS is essential for structural stability, environmental protection, and public safety, depending on riverbed conditions, geometry, sediment properties, flow velocity, and many other parameters. Feature reduction simplifies modeling, improves computational efficiency and focuses on key variables influencing scour depth. To enhance feature reduction, this paper presents a systematically validated framework using dimensional analysis and 13 state-of-the-art dimensionality reduction (DR) techniques. First, dimensional analysis identifies key parameters, followed by using various DR techniques to reduce features. By methodical testing and assessment, the best DR strategies are determined. A SWOT analysis is performed to evaluate the strengths, weaknesses, opportunities, and threats associated with each DR technique. According to SWOT analysis, subset regression analysis (SSR) emerges as the optimal DR technique compared to others, both quantitatively and qualitatively. According to quantitative evaluation, SSR outperforms Principal component analysis (PCA) (61.2 %) and Discriminant analysis (DA) (18.33 %), achieving the highest weighted mean percentage (70 %) among all. Furthermore, SSR excels by its ability to reduce computational complexity while maintaining high predictive accuracy. The efficiency and robustness of SSR are further supported by its low standard error of 0.21 and optimal selection of predictors, ensuring reduced computational demands without compromising accuracy. This strategic assessment provides in-depth insights into the technical effectiveness and practical viability of DR techniques. Through dimensional analysis and evaluating DR techniques, this study advances feature reduction methods for improved decision-making in fields like scouring around SDLS.