Edge-Cloud-Assisted Multivariate Time Series Data-Based VAR and Sequential Encoder–Decoder Framework for Multi-Disease Prediction

• Published in: Arabian Journal for Science and Engineering Vol. 50; no. 19; pp. 15729 - 15749
• Main Authors: Haq, Shamsul, Verma, Prabal
• Format: Journal Article
• Language: English
• Published: Berlin/Heidelberg Springer Berlin Heidelberg 01.10.2025; Springer Nature B.V
• Subjects: Algorithms; Artificial intelligence; Clinical outcomes; Cloud computing; Coders; Complex variables; Data analysis; Data processing; Decision making; Decoders; Deep learning; Disease; Edge computing; Efficiency; Electronic health records; Engineering; Epidemics; Forecasting; Humanities and Social Sciences; Integrated approach; Intervals; Machine learning; multidisciplinary; Multivariate analysis; Privacy; Research Article-Computer Engineering and Computer Science; Risk levels; Root-mean-square errors; Science; Time series; Vector autoregression (VAR); Sequential Encoder–Decoder; Cloud computing; Long Short-Term Memory networks (LSTM); Edge computing
• ISSN: 2193-567X; 1319-8025; 0377-9211; 2191-4281
• DOI: 10.1007/s13369-024-09898-3

Multivariate clinical time series data like demographic information, lifestyle factors, and health indicators for multi-outcomes are essential for capturing complex interdependencies between multiple variables and predicting outcomes simultaneously. Related to this context, this paper proposes a framework for analyzing multivariate time series data using the edge-cloud computing paradigm. In this framework setting, edge computing is utilized for the developing Hybrid Model consisting of “Vector Autoregression-Sequential Encoder and Sequential Encoder–Decoder” Models for forecasting outcomes. On the other hand, cloud computing plays the pivotal role of training Sequential Encoder (LSTM) and Sequential Decoders for complex large datasets for the updation of adaptive hybrid model after regular time intervals/horizons. These intervals or horizons are aligned with the conditions of risk levels, specifically reflecting the variations or toggling of risk states to accommodate window size for new features. Furthermore, experimental results demonstrate that the proposed framework outperforms existing models in terms of Mean Absolute Percentage Error (MAPE), Median Absolute Percentage Error (MdAPE), and Root-Mean-Square Error (RMSE) for diagnosing various diseases. The respective values of these statistical parameters are (0.200, 0.150, 0.012) for Diabetes, (0.061, 0.060, 0.010) for Hypertension, and (0.650, 0.600, 0.018) for heart attack, compared to other state-of-the-art models.