Analyzing IBM Watson’s Predictive Models for Heart Attacks, Heart Disease, and Heart Failure

Making comparisons of the features (age, chest pain, cardiac arrest, thalassemia, and blood pressure) present when looking at heart attacks, heart failure, and heart disease will help understand the correlation and any distinguishing features to help find any commonality between the three. Our hypothesis is that the diagnoses of heart attacks, failures, and diseases, all share some important variables, which may show presence of more than one of these in one patient, due to their linkage. In current work, we have compared different algorithms using IBM Watson to predict outcomes of heart attack datasets. When our heart attack data, collected from Kaggle, was run through IBM, accuracies as high as 86% from two datasets were achieved. First Dataset included 303 patients and the other held 1025 patients. In this study, multiple algorithms were simultaneously applied that included Logistic Regression, Snap Logistic Regression, and Gradient Boosting Classifiers. Continuing further on this investigation, another goal is to compare the similar features in heart attacks and diseases, which both can lead to heart failure. Using data from Kaggle, IBM Watson will be applied to compare previous results with the significant features and model accuracies with the results of the new data. Comparing the different variables show extremely similar features that may be used to diagnose each condition. The datasets also are compiled with data from different parts of the world, which will yield some variance in data points as cultural factors, such as food. These factors may vary in different parts of the globe to impact the model accuracies. A deeper AI/Machine Learning perspective will be presented.