<p>A non-communicable disease that affects the bones is called osteoporosis (OP). It results in altered bone microstructures, insufficient bone regeneration, and decreased bone mineral density (BMD). Until the disease progresses, individuals often don't realize they have it. Even with the widespread use of deep learning (DL) and machine learning (ML) algorithms, the early diagnosis of osteoporosis patients can still be improved. By automatically extracting hierarchical features from X-ray images through the application of a wide range of competitive ML (Gaussian Process (GP) kernel, SVM, XGBoost, Bagging), and DL models, we provide highly accurate and consistent predictions to accurately and early predict osteoporosis patients. In the proposed approach, we extracted and integrated significant features using three image feature extractors (LBP, CLBP, and HOG) with 95% PCA. Furthermore, we proposed upscaling (2 × &amp;4 ×) employing bicubic interpolation to enhance the image intensity. Subsequently, we applied the interpolated images to the suggested customized convolution neural network (CNN) model to improve the diagnostic performance. The proposed GP model outperformed other GP kernel and traditional ML models, achieving notable results with 67.74% accuracy, 61.97% precision, 92.52% recall, 74.12% F1-score, and a 67.77% AUC. To enhance osteoporosis classification, we developed a hybrid stacked CNN model, which demonstrated excellent performance on interpolated images. The proposed CNN model achieved state-of-the-art results, with 98% accuracy on the training set, and 95% accuracy on both the test and validation sets, significantly surpassing existing models. The numerical comparison shows the CNN model's superiority, with about a 30% increase in accuracy over the GP model. However, this research demonstrates the strong efficacy of the CNN model in handling complicated image data, making it a more sophisticated tool for osteoporosis detection.</p>