An Automatic System for Heart Disease Prediction using Perceptron Model and Gradient Descent Algorithm
V. Sahaya Sakila1, Akshat Dhiman2, Kanyush Mohapatra3, Panchal Raj Jagdishkumar4
1Akshat Dhiman*, Department of Computer Science and Engineering, SRM Institute of Science and Technology, Chennai, Tamil Nadu, India.
2Kanyush Mohapatra, Department of Computer Science and Engineering, SRM Institute of Science and Technology, Chennai, Tamil Nadu, India.
3Panchal Raj Jagdishkumar, Department of Computer Science and Engineering, SRM Institute of Science and Technology, Chennai, Tamil Nadu, India.
Manuscript received on September 22, 2019. | Revised Manuscript received on October 20, 2019. | Manuscript published on October 30, 2019. | PP: 1506-1509 | Volume-9 Issue-1, October 2019 | Retrieval Number: A1278109119/2019©BEIESP | DOI: 10.35940/ijeat.A1278.109119
Open Access | Ethics and Policies | Cite | Mendeley
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Abstract: Many deep learning neural network-based models had been proposed for the prediction of heart diseases but there is no model accessing all 13 features directly from the dataset & feeding them to the ‘Perceptron’. The model starts with certain weights and biases to train itself with the given dataset. The error is calculated in each epoch using the popular ‘gradient descent’ algorithm. Improper arrangement of neurons in a neural network can lead to overfitting or underfitting of the model which will gradually decrease the accuracy. But the perceptron model has only a single neuron performing all the classification and arrangement of data. The weights and biases in the model are updated to fit the model properly in a way that the equation can classify maximum number of data points. The learning algorithm of Perceptron model which calculates the result and performs binary classification is Σ_(i=0)^n wixi >= b. The weights and biases in the equation are changed repeatedly over epochs to decrease the error and fit the dataset with most efficient predicted output. Compared to the old MP Neuron model, the model is more effective with an accuracy of 98.349%.
Keywords: Gradient descent, Heart disease, Overfitting, Perceptron model, Underfitting.