A Genetic Algorithm for Feature Selection for Alzheimer’s Disease Detection Using a Deep Transfer Learning Approach

Alzheimer’s disease (AD) is one of the most common forms of neurodegenerative impairment. It is a progressive brain disorder affecting memory, thinking, and behaviour, ultimately leading to severe impairment and loss of independence. In predicting Alzheimer’s disease, it is widely recognized that handwriting is one of the first abilities affected by the onset of the disease. Most existing prediction systems focus on analyzing the dynamics of the handwriting process using online handwriting samples. However, these systems often fail to capture changes in handwritten characteristics’ shape, size, and thickness, which can indicate motor control alterations caused by neurodegenerative disorders. A previous study introduced a novel approach by combining shape and dynamic information to address this limitation. Synthetic colour images were generated from online handwriting samples, where each elementary trait’s colour encoded the associated dynamic information in the three RGB channels. Such a dataset was then used for classification through Deep Learning (DL). Moving from what was done, our study introduces a hybrid method, where Deep and Machine Learning (ML) techniques are used to implement a more powerful classification system to support the experts in diagnosing AD. Among the ML techniques considered, we performed two feature selections, one based on a recursive method and another on a genetic algorithm. Promising preliminary experimental results have confirmed the effectiveness of this proposed approach.