Essays on applications of deep learning for time series forecasting in intelligent decision support systems

Abstract

The accelerated evolution of artificial intelligence (AI) is driving a paradigm shift across diverse business landscapes, with growing efforts to integrate AI-driven solutions into the conventional decision support systems. However, practical application of these innovations requires the assurance of the AI algorithms' quality, reliability, and the aptitude for real-time decision-making. This prerequisite limits the practicality of many academic studies that focus solely on technology development. In this context, this dissertation presents the development of deep learning models applicable to intelligent decision support systems in healthcare and industrial information technology (IT) sectors and offers insights into their business value in terms of system adoption, utilization, and success. The first and second essays develop interpretable deep learning models for predicting hypnotic level and hypotension risk of patients during surgery and investigate their utility and impact in biometric signal monitoring systems. The third essay investigates deep learning models for detecting time series anomalies in IoT sensor data and studies its applicability in industrial monitoring systems.