TOPSIS-based Regression Algorithms Evaluation

Abstract

This paper developed a multi-criteria decision-making approach using the Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) to benchmark the regression alternatives. Regression is used in diverse fields to predict consumer behavior, analyze business profitability, assess risk, analyze automobile engine performance, predict biological system behavior, and analyze weather data. Each of these applications has its own set of concerns, resulting in various metrics utilizations or those of similar measures but with diverse preferences. Multi-criteria decision-making analyzes, compares, and ranks a set of alternatives utilizing mathematical and logical processes with a complicated and contradictory set of criteria. The developed approach established the weights, which were the core of the evaluation process, to various values to mimic and address the regression’s utilization in multiple applications with different concerns and using distinct datasets. The alternative judgment identified positive and negative ideal alternatives in the alternative space. The compared regression alternatives were scored and ranked based on their distance from these alternatives. The results showed that different preferences led to varying algorithm rankings, but top-ranked algorithms were distinguished using a specific dataset. Following that, using three datasets, namely Combined Cycle Power Plant, Real Estate, and Concrete, Voting using multiple classifiers (k-means-based classifiers) was the top-ranked in the Combined Cycle Power Plant and Real Estate datasets. In contrast, Decision Stump was the top-ranked in the Concrete dataset.