COMPARING SCALING METHODS AND DIMENSIONALITY REDUCTION TECHNIQUES: THEIR INFLUENCE ON CLASSIFICATION ALGORITHMS

Abstract

Machine learning in predictive analysis can be highly sensitive to its preprocessing methods, especially with scaling and dimensionality reduction. This study aims to create a unified and reproducible framework for supervised classification pipelines under varying preprocessing configurations. The framework is designed to integrate multiple classifiers with scaling methods and principal component analysis (PCA), enabling structured experimentation across different classification datasets. The models are evaluated using accuracy, precision, recall and F1-score, combined into a composite performance metric for comparison. The proposed approach is believed to produce a reproducible and extensible methodology for benchmarking classification pipelines in practical machine learning applications.