Optimized Machine Learning Framework for Osteosarcoma Diagnosis with Imaging and Survey Data

Osteosarcoma is an uncommon but deadly bone cancer that emphasizes how important it is to get a diagnosis early in order to improve patient outcomes. In order to increase diagnostic accuracy, this study offers an improved machine learning architecture that makes use of survey and imaging data. Using Principal Component Analysis (PCA) for feature extraction and sophisticated feature selection methods as Kruskal-Wallis, Chi-Squared, Fisher’s test, Student’s t-Test, and Recursive Feature Elimination (RFE), a strong preprocessing pipeline was created. Four machine learning methods were rigorously tested for their diagnostic performance: K-Nearest Neighbors (KNN), Random Forest, Support Vector Machine (SVM), and Naive Bayes. Random Forest and KNN outperformed SVM (89%) and Naive Bayes (88%) with the maximum accuracy of 94%. Evaluation criteria such as precision, recall, and F1-score further confirmed Random Forest’s dominance. The findings provide a potential method for diagnosing osteosarcoma by demonstrating the effectiveness of fusing feature engineering with machine learning techniques to create dependable automated cancer detection systems.