Comparative Analysis of Predictive Models for Post-Surgery Kyphosis Persistence: Using Machine Learning Techniques for Clinical Prognosis

Kyphosis, an abnormal curvature of the upper spine, frequently necessitates surgical treatment, particularly in adolescence. Yet, assessing the duration of the illness after treatment is difficult however essential for predicting the patient’s prognosis. The study evaluated the performance of various machine learning models on the Kyphosis classification task using K-fold cross-validation using 5 and 10 folds. The dataset was divided into 5 subsets with 5 folds and 10 subsets with 10 folds, allowing for accurate predictions on unseen data. K-Nearest Neighbours was the best-performing model, with an average cross-validation score of around 0.864. Random Forest and Support Vector Machine followed, achieving impressive scores of around 0.854 and 0.842, respectively. Naive Bayes showed slightly lesser performance, while Logistic Regression showed the lowest performance. In the 10 splits scenario, KNearest Neighbours continued to perform well, with an average cross-validation score of around 0.887. The results offer useful insights into how different models perform under various cross-validation methods, helping in the selection of suitable algorithms for predictive modelling. The study highlights KNearest Neighbours as the most reliable performer in both 5 splits and 10 splits. Basic ML algorithms-based models are competitive with deep learning models with accuracy rates ranging from 74% to 88%, offering a balance between performance and computational complexity. Further the integration of these models into clinical practice to provide dependable predictions regarding the persistence of Kyphosis after surgery, particularly in settings where computational resources are limited was suggested. This offers vital information for healthcare professionals to predict postoperative results with precision. Our research highlights the importance of utilizing machine learning as a key tool in tackling a variety of biological challenges.