GraphNet Framework for Schizophrenia Classification Using Sensor Data

A GraphNet framework is introduced for the classification of schizophrenia
using sensor-based motor activity data, aimed at modelling behavioural
patterns and discerning variations in movement activity between afflicted
people and control participants. The goal is to employ a structured graph
representation to show changes in temporal activity. The approach uses a
graph to show activity patterns, with nodes representing extracted
characteristics and edges showing how things change over time. This makes
it easy to understand how behaviour evolves. Before being turned into graph
inputs to capture both short-term and long-term dependencies, sensor data
are pre-processed, standardized, and broken up into useful time periods.
Feature representation makes it easier to see activity patterns and understand
how behaviour changes. The learning process identifies both individual and
relationship patterns within the data. The PSYKOSE dataset is used for
evaluation, and stratified K-fold validation is applied. The model achieves
an accuracy of 95.45%, precision of 100.00%, recall of 90.91%, and F1
score of 95.24%.