Determination of Significant Features for Building an Efficient Heart Disease Prediction System

Abstract

Heart diseases are responsible for the greatest number of deaths all over the world. These diseases are usually not detected in early stages as the cost of medical diagnostics is not affordable by a majority of the people. Research has shown that machine learning methods have a great capability to extract valuable information from the medical data. This information is used to build the prediction models which provide cost effective technological aid for a medical practitioner to detect the heart disease in early stages. However, the presence of some irrelevant and redundant features in medical data deteriorates the competence of the prediction system. This research was aimed to improve the accuracy of the existing methods by removing such features. In this study, brute force-based algorithm of feature selection was used to determine relevant significant features. After experimenting rigorously with 7528 possible combinations of features and 5 machine learning algorithms, 8 important features were identified. A prediction model was developed using these significant features. Accuracy of this model is experimentally calculated to be 86.4%which is higher than the results of existing studies. The prediction model proposed in this study shall help in predicting heart disease efficiently.

A hybrid approach for efficient outlier detection using supervised and unsupervised techniques

“…A hybrid technique that takes into account supervised and unsupervised ML algorithms is offered by the algorithm known as Multi-Model Approach for Outlier Detection (MMA-OD). The method design is in line with the assumption that supervised learning is superior for class label prediction while unsupervised learning has the capacity to accurately represent feature space [28]. These findings provide the basis for the suggested algorithm's mode of operation.…”

Enhancing Data Lake Management Systems with LDA Approach

“…It's worth noting that various topic modeling techniques exist, with LDA standing out as one of the most renowned and widely adopted methods for this purpose [24]. The versatility of LDA can be used to a wide range of document types, such as social media posts, policy documents, collections of news items, political science texts, software engineering documents, medical literature, linguistics research, and even short-form content like tweets [7]. Our proposal builds upon recent advancements in topic modeling techniques, particularly focusing on the execution of LDA within the context of data lakes.…”

“…In light of this, this study examines strategies for efficiently retrieve, analyze, organize, and uncover insights from text inputs. To put it briefly, topic modeling is an innovative and extremely successful technique for document classification automatically [7], comprehending vast amounts of textual data inside a large collection of unstructured documented data, and summarizing vast amounts of textual data.…”