Hybrid convolutional neural networks-support vector machine classifier with dropout for Javanese character recognition

This research paper explores the hybrid models for Javanese character recognition using 15600 characters gathered from digital and handwritten sources. The model combines merit of deep learning convolutional neural networks (CNN) to involve feature extraction a machine classifier support vector (SVM). dropout layer also manages overfitting problems enhances training accuracy. For evaluation purposes, we compared CNN with three different architectures multilayer perceptron (MLP) one two hidden layer(s). In this research, evaluated variants CNN-SVM on both accuracy classification time. experimental outcomes showed that performances all outperform MLP models. highest testing basic is 94.2% when 3 CNN. increment layers just slightly Furthermore, gained result 98.35% classifying data combining SVM classifier. We get can enhance results in recognition.