Hybrid Machine Learning Merging KNN and Naive Bayes Weighted Polling for Dry Beans Multiclass Classification

ASM Sc. J., 21(2), 2026

Published on October 14, 2026

https://doi.org/10.32802/asmscj.2026.66

Author: Farid Morsidi

Abstract

Multiclass classification is essential in machine learning as it allows systems to differentiate between three or more categories using intricate, multidimensional data. This paper examines the theoretical basis and real-world applications of multiclass classification, highlighting feature selection methods and the efficacy of classifiers. Essential principles, including significant inter -class variance, minimal inter-feature correlation, robust target relevance, and feature consistency, are examined for their contributions to discovering optimal predictors across various datasets. The experimentation is implemented with a dry bean dataset, utilising classifiers such as K-Nearest Neighbors (KNN) and Naive Bayes to distinguish bean types based on their morphological characteristics. KNN excels in basic pattern recognition, and Naive Bayes implements classification probabilistically, which is appropriate, widespread and accurate in high-dimensional environments. The study also stresses the importance of performance metrics such as accuracy, and dimensionality reduction as useful analytic tools. The method of feature selection and classifier combined achieved the accuracy measures from the lowest value of 85.67% to the highest value of 88.58%. Although these results look good for the automated classification of beans, performing thorough evaluati on with balanced metrics for all classes would increase the trust in the robustness of the technique, especially for the minority bean types.

Keywords: instance-based learning, KNN, machine learning, Naïve Bayes, probability classification