Machine Learning Approaches for Identifying Pathological Conditions in Phonocardiographic Data

A single PCG signal is subjected to 39 filters as part of the approach, producing 39 pro- cessed signals. Six signals are obtained from each of the first 39 signals using Multifractal Detrended Fluctuation Analysis (MFDFA), which then produces 240 signals for additional analysis One PCG signal is subjected to 39 filters as part of the approach, producing 39 processed signals. Multifractal Detrended Fluctuation Analysis (MFDFA) is then used, producing six signals from each of the first 39 signals, for a total of 240 signals for additional analysis. Modern algorithms such as Support Vec- tor Machines (SVM), Ensemble Tree Methods, k-Nearest Neighbors (kNN), and hybrid approaches are used for classification. The efficiency of the suggested approach is proven by extensive testing and evaluation, which successfully distinguishes between normal and pathological instances. Objective : Effective cardiac diagnosis depends on the accurate classification of PCG signals into normal and abnormal instances. This work offers a new approach to this kind of categorization, using a wide range of methods to produce trustworthy differentiation. Materials and Methods : In order to increase the number of signals by using subfrequency, digital filtering and multifractal detrended fluctuation analysis are applied. To increase discriminative power, 6,720 characteristics are derived from each distinct signal. These features capture key aspects of the phonocardiogram (PCG) signals, allowing for a thorough study. Modern classification methods like Support Vector Machines (SVM), Ensemble Tree Methods, k-Nearest Neighbors (kNN), and hybrid approaches are used after feature selection using Spearman correlation. Results : The ensemble tree approach performed exceptionally well for binary and multiclass classification in the results, achieving 100% accuracy on a range of metrics when using MFDFA and subfrequency approaches. Conclusion : This paper provides a comprehensive account of the methodology, including detailed explanations of the classification process, feature extraction techniques, and evaluation of results.