CNN feature and classifier fusion on novel transformed image dataset for dysgraphia diagnosis in children

Dysgraphia is a neurological disorder that hinders the acquisition process of normal writing skills in children, resulting poor abilities. Poor or underdeveloped children can negatively impact their self-confidence and academic growth. This work proposes various machine learning methods, including transfer via fine-tuning, feature extraction, ensembles deep convolutional neural network (CNN) models, fusion CNN features, to develop preliminary dysgraphia diagnosis system based on handwritten images. In this work, an existing online dataset converted into images, encompassing tasks. Transfer applied using pre-trained DenseNet201 four distinct models separately trained word, pseudoword, difficult sentence Soft voting hard strategies are employed ensemble these models. The used for extraction from each task-specific image data. extracted features then fused different combinations. Three algorithms support vector (SVM), AdaBoost, Random forest assess performance features. Among SVM word data achieved highest accuracy 91.7%. case learning, soft CNNs 90.4%. approach substantially improved classification accuracy, with task specific-data achieving 97.3%. surpasses state-of-the-art methods by 16%.