Knee joint disorders are a significant health concern globally, with conditions like osteoarthritis affecting millions of individuals. In the current scenario, early detection and accurate classification of knee health conditions are crucial for effective management and treatment. A hierarchical frequency zone clustering (HFZC) method based on the Fourier-Bessel Series Expansion Empirical Wavelet Transform (FBSE-EWT) is shown in this paper as a way to look at vibroarthrographic (VAG) signals and put knee joint problems into different groups. FBSE-EWT is used to split VAG signals into 50 sub-bands for this method. The FBSE-EWT decomposed sub-bands have shown different filter bands for different signals. Due to the unfixed frequency width of a specific sub-band of various signals, determining spectral features from these sub-bands and comparing them has resulted in incorrect VAG signal interpretation. To address this issue and enhance the accuracy of classification, we suggest implementing a hierarchical frequency zone clustering (HFZC) approach. In HFZC, we form fixed frequency clusters from the decomposed FBSE-EWT sub-bands. We compute and frame spectral descriptors as feature vectors for each of these clusters, as well as their variability measures. The recursive feature elimination (RFE) method on the full set of spectral descriptors and their variability measures to find the most important features that can tell the difference between the normal and abnormal classes were used. Investigated the ability of these features for VAG signal classification using prominent classifiers such as random forest (RF), XG boost (XGB), extremely randomized tree (ERT), light gradient boosting machine (LGBM), gradient boosting (GB), ADA boost (AB), and an ensemble of these classifiers (ENSC). The proposed methodology, utilizing an ensemble classifier, demonstrated the highest accuracy of 94.38%, sensitivity of 92.11%, specificity of 96.08%, and an AUROC of 0.94.
