Heart Beat Sound Signals Classification Using Spectral Roll Off

cardiovascular diseases, abnormalities or dysfunctions related to heart or blood vessels may occur due to a number of reasons i.e., anxiety, stress, higher blood pressure, cholesterol, diabetes, alcohol consumption, and smoking. Some cardiovascular diseases also involve the presence of irregular heartbeats as either symptoms or indicators of other underlying conditions. The proposed research work aims to identify and classify a patient's heartbeat as irregular or regular based on the sample data inputs. However, the sample data consists of unbalanced classes i.e., sample size of irregular heartbeats. This issue has been addressed through SMOTE up-sampling technique. Another problem exists while trying to achieve the end goal of identifying a patient's heartbeats as regular or irregular, that is, the presence of a high number of features i.e., around one lakh ten thousand. Due to this, the classification model becomes complex and analyzing data becomes inefficient and complex. This also creates the problem of the curse of dimensionality which, here, is addressed through SPECTRAL ROLL OFF of featured extraction. Post this, features will be further reduced to a smaller number using MEAN SPECTRAL ROLL OFF of feature extraction. Once the features have been reduced, analysis is performed. Through this analysis, logistic regression predicts the incoming patient's heartbeats as regular or irregular. Through the performance matrices, the result's accuracy is declared.