Leukocoria Identification: A 5-Fold Cross Validation CNN and Adaboost Hybrid Approach
Indah Purnama Sari, Al-Khowarizmi Al-Khowarizmi, Fanny Ramadhani, Andy Satria, Oris Krianto Sulaiman
Abstract
Leukocoria, which is distinguished by an unusual white reflection in the pupil, is an important sign of several eye conditions. Early detection and correct alleviation of leukocoria are critical for prompt diagnosis and therapy. In this article, we propose a hybrid method for leukocoria classification utilizing Convolutional Neural Networks (CNN) and Adaptive Boosting (AdaBoost). The performance of the suggested strategy is assessed using the 5-Fold Cross Validation method in the proposed work. In order to increase classification accuracy, CNN is utilized for feature extraction and AdaBoost is employed as a boosting technique. The experimental results reveal that the suggested combined strategy is more effective in relieving leukocoria than the CNN or AdaBoost single models. In comparison to the conventional CNN technique, the performance results of the model utilizing the CNN + Adaboost method with 5-Fold Cross Validation offer better and superior outcomes in diagnosing leukocoria. Results from the CNN + Adaboost model using 5-Fold Cross Validation show very high accuracy values of 95%, precision values of 95%, recall values of 95%, and Fl scores of 94%.