Osteoporosis Detection Using a Combination of Recursive Feature Elimination and Naive Bayes Classifier with Rule-Based Chatbot Testing

Abstract

Osteoporosis is a condition characterized by reduced bone mass and density, increasing the risk of fractures. Early detection relies on patient awareness and proactive health management. Despite advances in technology, patient independence and awareness remain critical for early diagnosis. A rule-based chatbot tool can assist by helping patients screen their bone health. The chatbot provides automated recommendations, offering an alternative to traditional hospital visits. This study presents a rule-based chatbot designed to detect osteoporosis, using Recursive Feature Elimination (RFE) combined with the Naïve Bayes Classifier (NBC). Machine learning is integrated to enhance the chatbot's ability to identify early signs of osteoporosis. The model’s performance is compared to other feature selection methods, such as Principal Component Analysis (PCA), and machine learning algorithms like Deep Learning, Support Vector Machine (SVM), and Logistic Regression. The dataset used includes public data sets for training and validation, as well as data from the Yogyakarta Health Office for predictions. Research phases include normalization, data encoding, feature selection, training, validation, and prediction. The chatbot implements text preprocessing techniques, such as tokenization, stop word removal, and feature extraction, alongside normalization and encoding of numeric data. The prediction stage determines if the patient has a positive or negative osteoporosis status. Validation results show the RFE-NBC model is particularly effective for osteoporosis detection, offering a balanced performance in identifying both positive and negative cases. Additionally, this model served as the foundation for creating a rule-based chatbot designed to detect osteoporosis. Based on the set of testing metrics using chatbot, the model demonstrates strong overall performance, with a good balance between identifying positive and negative instances.