Ensemble learning for biomedical signal classification: a high-accuracy framework using spectrograms from percussion and palpation.

Accurate classification of biomedical signals is crucial for advancing non-invasive diagnostic methods, particularly for identifying gastrointestinal and related medical conditions where conventional techniques often fall short. An ensemble learning framework was developed by integrating Random Forest, Support Vector Machines (SVM), and Convolutional Neural Networks (CNN) to classify spectrogram images generated from percussion and palpation signals. The framework employs Short-Time Fourier Transform (STFT) to extract spectral and temporal information, enabling accurate signal processing and classification into distinct anatomical regions. The ensemble model combines the strengths of its components: Random Forest mitigates overfitting, SVM handles high-dimensional data, and CNN extracts spatial features within a robust preprocessing pipeline to ensure data consistency. By achieving a classification accuracy of 95.4%, the ensemble framework outperformed traditional classifiers in capturing subtle diagnostic variations. This method offers a robust solution for biomedical signal classification and has potential applications in clinical diagnostics. Future research directions include real-time clinical integration and multi-modal data incorporation to further enhance its applicability.