Trustworthy AI in Healthcare: Insights, Challenges, and the Significance of Overfitting in Predicting Mental Health

Abstract

The rapid integration of artificial intelligence (AI) into medical informatics, particularly in the context of mental health data, can bring about significant transformations in healthcare decision-support systems. However, ensuring that AI gains widespread acceptance and is regarded as reliable in healthcare requires addressing critical issues concerning its robustness, fairness, and privacy. This chapter presents a comprehensive study that delves into the urgent need for dependable AI in medical informatics, explicitly focusing on collecting mental health data using sensors. The authors put forth a methodological framework combining cutting-edge AI techniques, leveraging deep learning models such as recurrent neural networks (RNN), including variants like LSTM and GRU, and ensemble techniques like random forest, AdaBoost, and XGBoost. Through a series of experiments involving healthcare decision support systems, the authors underscore the pivotal role of model overfitting in establishing trustworthy AI systems.