Schizophrenia is a severe mental illness affecting over 20 million people worldwide, significantly impairing quality of life and daily functioning. Current diagnostic methods rely heavily on subjective assessments and interactions between doctors and patients, leaving room for potential misdiagnoses. Recent advancements in technology have introduced non-invasive, fast, and user-friendly approaches, such as machine learning, to support psychiatric diagnosis. In this study, spectral features extracted from speech samples of individuals with and without schizophrenia were analyzed. Using an ensemble bagged tree model, we achieved an accuracy of 96.3%, a sensitivity of 94.6%, and an F1-score of 95.4%. These results highlight the potential of speech-based machine learning models as effective tools for aiding schizophrenia diagnosis.

Neural networks have achieved remarkable success in various applications such as image classification, speech recognition, and natural language processing. However, the growing size of neural networks poses significant challenges in terms of memory usage, computational cost, and deployment on resource-constrained devices. Pruning is a popular technique to reduce the complexity of neural networks by removing unnecessary connections, neurons, or filters. In this paper, we present novel pruning algorithms that can reduce the number of parameters in neural networks by up to 98% without sacrificing accuracy. This is done by scaling the pruning rate of the models to the size of the model and scheduling the pruning to execute throughout the training of the model. Code related to this work is openly available.