Abstract |
: |
Diabetes produces diabetic retinopathy, which damages the retina and impairs eyesight. Untreated, it can cause blindness. DR is irreversible; therapy just preserves vision. Early DR detection and treatment reduce visual loss. Manual diagnosis of DR retinal fundus images by ophthalmologists is time-, effort-, and cost-intensive and prone to misdiagnosis. Deep learning has improved numerous sectors, including medical image processing and categorization. In 2015, 2.6 million people were visually impaired or blind due to diabetic retinopathy, and 3.2 million by 2020. Diabetic retinopathy is expected to diminish in high-income countries, although early diagnosis and treatment are still important in low- and middle-income countries. This research work finds that automated screening and grading of diabetic retinopathy reduces human and saves time and resources by using machine learning using image enhancement techniques.Classification via regression's model creation takes 1.48 seconds. Random Sub Space takes 0.16 seconds to build its model. The Multiclass classifiers have the highest accuracy at 74.89%. The Bagging has minimum model accuracy of 66.1%.The multi class classifier produces 0.76 of precision which is maximum precision of selected classifiers. Bagging produces 0.66 precision, the minimum for selected models.The multiclass classifier has 0.75 recall, the highest among selected classifiers. Bagging produces 0.66 recall, the minimum for selected models.Multi Class Classifier gives kappa, F-Measurer, and MCC values of 0.5. The Regression Classification yields 0.35 kappa, 0.68 F-Measure, and 0.35 MCC.Multiclass classifier produces 0.5 kappa, the highest of selected classifiers. Bagging produces 0.32 kappa, the minimum of selected models. Multi-class classifier produces 0.75 F-Measure, the maximum of selected classifiers. Bagging produces 0.66 F-Measure, the minimum for selected models.Multiclass classifier produces 0.51 MCC, the maximum of selected classifiers. Bagging produces 0.32 MCC, the minimum for selected models. Multi Class Classifier's ROC and PRC are 0.83. Classification via regression yields 0.73 ROC and 0.71 PRC. The multiclass classifier has the highest ROC (0.83). Bagging, Random Sub Space, and Regression Classification all produce 0.73 ROC. Multiclass classifier produces 0.83 PRC, the highest of selected classifiers. Random Sub Space Bagging produces 0.72 PRC. This work explores the multi class classifier shows best accuracy compare with other models and it gives low deviations. |