Optimizing PSO for classification: comparison of Naïve Bayes and C4.5 for osteoporosis prediction

Abstract

Osteoporosis is a medical disease marked by a reduction in bone density, which significantly increases the risk of fractures. Osteoporosis patients do not always exhibit symptoms and because current diagnostic techniques have limitations, early detection is frequently needed. The osteoporosis dataset consists of 1.958 records each containing 15 regular attributes and 1 special attribute as the label.  The attribute represented as “1” for the presence of osteoporosis and “0” for its absence. The primary objective is to predict an individual’s risk of developing osteoporosis, including age, gender, bone density, lifestyle factor, medical history, and nutritional intake of calcium and vitamin D. To achieve this, Naïve Bayes and C4.5 has been employed. PSO is employed to identify the most relevant features, thereby optimizing the efficiency and accuracy of the classification models. The initial step in data preprocessing involved handling missing values to ensure data integrity. After implementing PSO, Naïve bayes improved from 82,65% to 83,67%, while C4.5 exhibited an even greater increase, rising from 91,07% to 96,17%. PSO significantly optimizes model, with the most improvement in C4.5. PSO proves to be a valuable tool for feature selection. Age and Hormonal Change emerged as important for both models. Furthermore, Physical Activity and Calcium Intake, which despite having varying levels of influence, were consistently considered relevant.  By focusing on these significant attributes, enables us more effectively monitor and recognize early signs of osteoporosis. Identifying individuals at high risk, more effective early detection and intervention, improving the potential for timely management and prevention.