Machine Learning for Kyphosis Disease Classification

Kyphosis, a spinal disorder characterized by an abnormal curvature of the upper spine, often requires accurate diagnosis for timely intervention. In this study, machine learning algorithms—Random Forest, Adaboost, Naive Bayes, and Logistic Regression—were employed to classify the presence of kyphosis disease and predict the necessity of surgical intervention based on patient data. The dataset comprised a collection of clinical attributes such as age, vertebral column measurements, and the presence or absence of the kyphosis condition post-surgery. Each algorithm underwent rigorous training, validation, and testing processes using this dataset to ascertain its efficiency in disease classification. Results indicated that all four algorithms demonstrated varying degrees of efficacy in identifying the presence of kyphosis. Random Forest and Adaboost exhibited higher accuracy and robustness in distinguishing cases where the disease was present or absent, surpassing Naive Bayes and Logistic Regression. Additionally, they provided valuable insights into the key attributes contributing to accurate predictions. The findings suggest promising potential for machine learning algorithms, particularly Random Forest and Adaboost, in aiding medical practitioners in early diagnosis and decision-making regarding surgical intervention for kyphosis patients. Further refinement and validation of these models using larger and diverse datasets could significantly enhance their applicability in clinical settings. 

 Keywords: Random forest,  Adaboost,  Naive bayes, Logistic regression.