Comparative Analysis of Convolutional Neural Network Transfer Learning Models for Predicting Learning Disability

Abstract

A sizeable fraction of the population is impacted by learning difficulties, making early detection essential for successful intervention. In order to predict learning difficulties, convolutional neural network (CNN) transfer learning models are compared in this study. We examined the efficacy of four well-known CNN architectures: AlexNet, VGG16, ResNet50, and Inception, using handwritten image data. Our research shows that using pre-trained CNN models for learning disability prediction using transfer learning is effective. ResNet50 regularly beat other models in a variety of evaluation parameters, demonstrating its efficacy in correctly diagnosing people with learning difficulties. While Inception displayed somewhat poorer accuracy, AlexNet and VGG16 demonstrated competitive performance. The enhanced functionality of ResNet50 highlights the significance of skip connections and deeper structures in identifying intricate dataset patterns linked to learning impairments. This comparative study offers significant insights into the use of CNN transfer learning models for predictive analytics in learning disabilities. The results underscore the importance for early detection and intervention, highlighting the promise of machine learning methods in overcoming diagnostic and management challenges associated with learning disabilities.