Optimizing Supervised Classification Algorithms for Noisy, Real-World Data

Abstract

Supervised Machine Learning (SML) involves the identification of algorithms that learn from externally provided data instances to generate generalized models, which are then used to make predictions about future data. Among the various tasks performed by intelligent systems, supervised classification is one of the most commonly executed. This paper outlines several Supervised Machine Learning (ML) classification techniques, compares different supervised learning algorithms, and identifies the most effective classification method based on dataset characteristics, including the number of instances and features. Seven distinct machine learning algorithms were examined: Decision Table, Random Forest (RF), Naïve Bayes (NB), Support Vector Machine (SVM), Neural Networks (Perceptron), JRip, and Decision Tree (J48), utilizing the Waikato Environment for Knowledge Analysis (WEKA) tool. For the analysis, a Diabetes dataset containing 786 instances with eight independent attributes and one dependent attribute was used for classification. The findings reveal that SVM achieved the highest precision and accuracy. Naïve Bayes and Random Forest algorithms followed as the next most accurate methods after SVM. The study highlights that the time required to develop a model, as well as the precision (accuracy), is a critical factor, while Kappa statistic and Mean Absolute Error (MAE) also play significant roles. Thus, for effective supervised predictive machine learning, algorithms must prioritize precision, accuracy, and minimal error to have supervised predictive machine learning.