Learned prediction of cholesterol and glucose using ARIMA and LSTM models – A comparison

• Published in: Results in control and optimization Vol. 14; p. 100362
• Main Authors: Krishnamoorthy, Umapathi, Karthika, V, Mathumitha, M K, Panchal, Hitesh, Jatti, Vijay Kumar S, Kumar, Abhinav
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.03.2024; Elsevier
• Subjects: Autoregressive integrated moving average (ARIMA); Cholesterol prediction; Glucose prediction; Long short-term memory (LSTM); Machine learning; Non-invasive; Cholesterol prediction; Autoregressive integrated moving average (ARIMA); Non-invasive; Glucose prediction; Long short-term memory (LSTM); Machine learning
• ISSN: 2666-7207; 2666-7207
• DOI: 10.1016/j.rico.2023.100362

Painless remedies and diagnosis have become the primary objective in medical sciences as diseases shoot up. This research addresses the pressing need for cost-effective, non-invasive, and real-time monitoring solutions for glucose and cholesterol levels to reduce mortality rates associated with diabetes and cardiovascular diseases. Diabetes and cardiovascular diseases are significant global health challenges, affecting millions of individuals, particularly in low to middle-income countries. The inconvenience and discomfort of traditional blood tests have often resulted in delayed testing and monitoring, making it imperative to develop alternative methods for early disease detection and continuous monitoring. The primary objective of this research is to create a non-invasive glucose and cholesterol monitoring and forecasting system that leverages Artificial Intelligence and Cloud computing. The specific focus is comparing the prediction efficiency of two models: the Auto-Regressive Integrated Moving Average (ARIMA) model and the Long Short-Term Memory (LSTM) model. The research methodology involves the development of a hardware prototype for the non-invasive measurement of glucose and cholesterol, employing Near Infra-Red (NIR) sensors. Data on glucose and cholesterol levels are collected and stored from patients over a period of one month. The collected data is used to train both the ARIMA and LSTM models. The models are evaluated using Root Mean Square Error (RMSE), and their prediction accuracy is compared. The study findings reveal that the ARIMA model surpasses the LSTM model in predicting glucose and cholesterol levels. This is supported by a lower RMSE of the ARIMA model (about 71.7 % less for Glucose and 50.3 % less for Cholesterol) compared to the LSTM model, indicating higher prediction accuracy. Additionally, the hardware prototype facilitates non-invasive measurement and forecasting, offering a promising solution for improved early diagnosis and monitoring. The research highlights the potential of ARIMA and LSTM models in healthcare and diagnostic applications. The superior performance of the ARIMA model in predicting glucose and cholesterol levels positions it as a valuable tool for real-time applications. This study provides a unified solution for non-invasive measurement and forecasting, aiming to enhance early diagnosis and monitoring of diabetes and cardiovascular risk.