Home // ACCSE 2020, The Fifth International Conference on Advances in Computation, Communications and Services // View article

Prediction of Diabetic Retinopathy and Classifiers Sensitivity Analysis

Authors:
Khaled Almustafa

Keywords: Diabetic Retinopathy, Stochastic Gradient Decent, Logistic Regression, Multilayer Perceptron, Classification, Prediction, Feature Extraction, Sensitivity Analysis.

Abstract:
Many eye diseases, such as Diabetic Retinopathy (DR), can lead to blindness without early clinical diagnosis, and it is extremely important to take the necessary measures before it is too late. A reliable system to detect such a disease in an early stage would be a great addition to the health care providers. In this paper, a comparative analysis of different classifiers was done for the classification of the DR dataset using different machine learning classification algorithms, such as, Naïve Bayes, J48, Random Forest (RF), Stochastic Gradient Decent (SGD), Logistic Regression (LR), Multilayer Perceptron (MP), Simple Logistic (SL) and Logistic Model Tree (LMT) classifiers, and to measure the classification accuracy, the Area Under Curve (ROC), Mean Absolute Error (MAE) and Square Root Mean Square Error (RMSE) for classifying the DR dataset. The results showed that the Logistic Regression classifier outperformed all other classifiers in the classification of the DR dataset for a classification accuracy of 74.8914%, area under curve ROC = 0.831, and RMSE = 0.4061. Then a sensitivity analysis for MP classifier was investigated in term of changing its learning rate. Also, a feature extraction method was performed on LR, MP, SL and LMT classifiers to evaluate the classification performance after selecting the relevant attributes, and the results showed that an accuracy of 72.3719% can be obtained to predict a DR case using Multilayer Perceptron by only applying a combination of up to 8 attributes instead of 19 attributes of the full dataset.

Pages: 1 to 7

Copyright: Copyright (c) IARIA, 2020

Publication date: September 27, 2020

Published in: conference

ISSN: 2519-8459

ISBN: 978-1-61208-810-5

Location: Lisbon, Portugal

Dates: from September 27, 2020 to October 1, 2020