Knee osteoporosis is a structural disorder characterized by decreased bone mineral density, which increases the risk of fractures and significant morbidity. Early and accurate diagnosis is crucial for timely intervention and treatment, as traditional diagnostic methods rely on manual interpretation of X-ray images, which is time-consuming and subject to inter-examiner variability. This research presents an automated system for classifying and detecting knee osteoporosis that leverages a novel methodology for extracting, selecting, and classifying features via machine learning. We proposed a new hybrid multi-modal feature fusion technique, Hybrid Multi-Modal Feature Fusion (HMMFF), that integrates traditional feature extraction methods such as GLCM, HOG, and ORB with deep learning features from models including YOLOv11, VGG-16, and ResNet50. To improve the classification efficiency of the proposed system, feature selection was performed using the harmony search optimization algorithm. We also proposed a new hybrid feature selection method, the Hybrid Fuzzy Harmony Search Algorithm (HFHSA), that combines harmony search with Gath-Geva fuzzy clustering. The selected features were then fed into the proposed machine learning classifiers and evaluated. The results indicated that the XGBoost model achieved the highest accuracy rate, reaching 93.58% when using the Harmony Search algorithm alone for feature selection, While the Gradient Boosting model achieved the highest accuracy compared to the other models rate rose to 94.87% when using the HFHSA technique for feature selection. These results confirm the effectiveness of applying the HMMFF technique for feature extraction, along with feature selection using the HFHSA technique, in identifying important features, thereby providing an accurate diagnostic tool that enhances early detection of osteoporosis.