Hyper parameter tuning based gradient boosting algorithm for detection of diabetic retinopathy: an analytical review

Abstract

The pipelines of approaches for classifying diabetic retinopathy were examined in this study. The effort entails developing appropriate transformations and estimators that can be used to automate the process of diabetic retinopathy detection. The segmentation of the blood vessels was done using a hybrid algorithm that uses Otsu and median filter to get the region of interest. Further, ten classifiers were investigated in order to develop an automated pipeline for diabetic retinopathy detection. The ten classifiers were reviewed based on earlier work in a similar setting and on an exploration of new ways for identifying diabetic retinopathy. To overcome the challenge of low volume of dataset, data argumentation was done so that a generic classifier can be configured. Extensive hyper parameter tuning was performed, and it was shown that the gradient boosting approach is the most stable technique for detecting diabetic retinopathy. This was validated using a 10K fold cross validation method on many metrics (accuracy, recall, precision, and v-measure score). Hyper-parameter tuning helped in achieving accuracy of 0.96.