What if the earliest signs of cognitive decline could be detected through subtle patterns in everyday life?
This question shaped the research journey of SMU Master of IT in Business (MITB) and PhD in Computer Science alumnus Teh Seng Khoon. What began as a capstone project during his postgraduate studies evolved into years of research exploring how artificial intelligence (AI) and data science can identify early indicators of cognitive decline, a journey made possible by the MITB programme’s interdisciplinary training and research opportunities.
As populations age around the world, the need for early detection of cognitive decline has become increasingly important. Subtle changes in movement, sleep, and daily routines can emerge long before clinical symptoms appear, making early detection crucial for timely intervention and better long-term outcomes.
As part of his MITB capstone project, Seng Khoon worked on an in-home sensor study that explored how data science could address ageing-related healthcare challenges. With the introduction of a part-time PhD pathway at SMU, he continued pursuing the research alongside his professional work, transforming his capstone project into a long-term research journey on AI-driven cognitive health detection.
Understanding behavioural signals through data
To better understand how early detection could be achieved in practice, Seng Khoon first examined how digital health technologies were being used to study cognitive decline.
In a systematic review and meta-analysis published in the IEEE Journal of Biomedical and Health Informatics, he analysed existing studies that used digital biomarkers — data collected from sensors tracking behaviours such as movement, sleep, or daily activities — to identify mild cognitive impairment (MCI).
The review revealed that many studies still relied heavily on traditional machine learning techniques such as support vector machines and decision trees, despite the growing complexity of behavioural data being collected.
The study also highlighted the challenges of interpreting behavioural signals in real-world environments. For instance, movement-related data could point to both frailty and cognitive decline, making it difficult to distinguish between different health conditions.
These findings helped shape the direction of Seng Khoon’s subsequent research: developing AI models capable of extracting meaningful insights from complex behavioural data collected in everyday living environments.
Reflecting on his journey, Seng Khoon shares, “The diverse skillset gained from the MITB programme enabled me to handle the data analysis independently during my capstone project, providing a strong foundation for more advanced research during my PhD.”
For Seng Khoon (3rd from right), the MITB programme provided a strong foundation for advanced research in cognitive health detection, using sensors to collect behavioural data from everyday living environments.
Building AI models for real-world healthcare settings
In a second study published in Expert Systems With Applications, Seng Khoon developed a neural-network model designed for real-world healthcare datasets.
Built on adaptive resonance theory, it uses a self-organising neural network architecture known as fuzzy Adaptive Resonance Associate Map (ARAM) to identify behavioural patterns associated with MCI from data collected by in-home sensors.
This approach proved particularly effective for healthcare applications, where datasets tend to be relatively small and interpretability is critical. Drawing on the interdisciplinary training he received through the MITB programme – spanning machine learning, data preparation, and business workflow analysis – Seng Khoon was able to work across the technical and practical dimensions of healthcare data analysis.
“In healthcare, it is important to achieve high sensitivity and specificity while maintaining interpretability, so clinicians can trust the model,” Seng Khoon elaborates. “That guided our exploration of self-organising neural network architectures for MCI detection.”
The system analyses behavioural signals from ambient home sensors, including motion, door, bed, and wearables, collected over periods of up to three years. It identifies subtle patterns linked to MCI, such as lower movement activity and greater variability in sleep patterns, which may be difficult to detect clinically.
These findings highlight how AI can surface behavioural signals that might otherwise go unnoticed. Building on this work, Seng Khoon has further refined self-organising neural network architectures for digital biomarker analysis, with his latest research currently under peer review.
Advancing the future of cognitive health monitoring
Looking ahead, Seng Khoon believes future advancements in AI and sensor technologies could significantly improve the practicality of cognitive health monitoring systems.
Currently, most AI models using home sensor data are limited to tracking a single resident within a household, restricting their deployment in more complex real-world living environments. Future research will focus on developing algorithms capable of reliably analysing behavioural data from households with multiple occupants.
If successful, such systems could eventually become part of everyday homes, continuously monitoring behavioural patterns and detecting early signs of cognitive decline long before symptoms become clinically visible.
Enabling tech professionals to create meaningful impact with the SMU MITB
For Seng Khoon, his journey reflects the impact that the SMU MITB programme is designed to create: equipping professionals with interdisciplinary skills to tackle complex real-world challenges.
By combining technology, analytics, and real-world application, the MITB programme empowers students to move beyond technical theory and create solutions with tangible societal impact — whether in healthcare, business, or emerging digital industries.
