Heart Disease Prediction using Machine learning algorithms

Abstract

We all know that heart is the organ which pumps blood throughout our body. It is the primary organ of our circulatory system. If it fails to work properly, then the brain and various other organs will stop working, and within a few seconds the person will die. In our project Our Aim is to build the best predictive model using various machine learning algorithms or neural networks which gives the highest accuracy in predicting whether a patient has a heart disease or not using the patient’s data. We designed two approaches: In our first approach we are using Cleveland Heart Disease dataset which is a standardized dataset taken from the University of California Irvine (UCI) Repository. This dataset contains a total of 76 attributes but all previous research and published experiments refer to using a subset of 14 of them. we implemented some of the machine learning algorithms like Random Forest, Naïve Bayes, Decision trees, KNN. So far Random Forest Classifier has given most significant results with accuracy of 93.84%, this was achieved by the tuning of the hyperparameters of the algorithms with metric='Manhattan', n_neighbors=13, weights='distance', n_jobs=-1 at last these parameters values were used for extraction of the results. Now here comes our second method, we introduced heart disease detection based on heart sounds. The proposed method employs three successive stages, like spectrogram generation, deep feature extraction, and classification. In the spectrogram generation stage, the heart sounds are converted to spectrogram images by using time–frequency transformation. Our method also aims to classify an abnormal heartbeat with a normal heartbeat based on audio data recorded from a stethoscope. Audio data used is of wav type (Waveform Audio File Format). The classification is conducted using Convolutional Neural Network. Here we also studied Mel spectrogram and MFCC. In this, The epoch values used were 100,150, 200, 250 and 300. The best results were obtained with 300 epochs at 0.001 learning rate applied on batch size of 128. The training accuracy is 89.73%, while the testing accuracy rate is 82%.