Artificial Neural Networks with Enhanced Nearest Neighbour algorithm for Heart Disease Prediction

Abstract

Heart failure is the root cause of the increased mortality rate happening now in the current scenario. Many health care services have been investing in the research forefronts to improvise the predictive methodology of cardio arrests and cardio related diseases to avoid this major health hazard. The improvised predictive methodology by the research forefronts with timely response helps to provide proper medical aid before the condition gets severe. The forecast methods use artificial intelligence which includes the neural networks and fuzzy techniques. The neural network algorithm makes use of the datasets and predicts the future datasets under various conditions. The fuzzy technique implements the data sets performed by the neural algorithms provides automated training to the system which in turn makes the system reliable to device any decisions on their own under any critical circumstances. To contribute in providing the solution, this paper analyzes the invisible medical information of medications in-taken for various medical issues and their effects on the patients along with the patient’s medical history are taken as dataset inputs to the classification with association along with the nearest neighbor which process the possibilities and facts that bares the key reason for cardiac arrest. To avoid critical circumstances in the last minute of cardiac arrest, the neural algorithms forecast the cardiac related hazards on time which could help them to survive even in critical condition with proper medication and intensive care. There are 14 Attributes such as gender, cardiac acute type, pressure level, body glucose level, age factor, cholesterol and body fat contents, BMI, medications are the key factors taken as the primary source for the input datasets. The interpreted data sets are analyzed and visualized as patterns. The nearest neighbor algorithm is enhanced for recognizing the patterns and grouping the similarities and dissimilarities which later formed as data subsets for the forecast. To improvise the efficiency of results 14 attributes are taken as input datasets and by the process of neural algorithm, it is then reduced to 6 attributes results 99.1% accuracy with 0.9% of noise figures.