TY - CHAP

T1 - Ensemble Method Based on Markov Models for Time Series Forecasting

AU - Belousova, Ekaterina E.

AU - Rakitskiy, Anton

N1 - Conference code: 5

PY - 2025

Y1 - 2025

N2 - Markov chains are a powerful mathematical tool widely used for modeling stochastic processes. This paper provides an overview of the Markov chains concept and their application in predicting time series data. While traditional Markov models rely on first-order dependencies, higher-order chains capture complex temporal patterns. Using ensemble methods instead of simple model can bring significant advantages, especially when working with complex real-world data. Ensembles can demonstrate greater robustness to outliers and noise and, by combining the models, achieve better generalization capabilities. By combining Markov models of varying orders, ensembles mitigate noise sensitivity and enhance generalization capabilities. This work introduces a novel ensemble method where weighted coefficients combine predictions from multiple Markov chains. The weights are assigned based on the concept of the R-measure, a theoretical framework inspired by universal coding principles, which serves as a consistent probability estimator for stationary and ergodic processes. Experimental results on real dataset are provided, demonstrating the potential of proposed method. This research underscores the potential of ensemble-based Markov models as a scalable and reliable tool for real-world forecasting tasks.

AB - Markov chains are a powerful mathematical tool widely used for modeling stochastic processes. This paper provides an overview of the Markov chains concept and their application in predicting time series data. While traditional Markov models rely on first-order dependencies, higher-order chains capture complex temporal patterns. Using ensemble methods instead of simple model can bring significant advantages, especially when working with complex real-world data. Ensembles can demonstrate greater robustness to outliers and noise and, by combining the models, achieve better generalization capabilities. By combining Markov models of varying orders, ensembles mitigate noise sensitivity and enhance generalization capabilities. This work introduces a novel ensemble method where weighted coefficients combine predictions from multiple Markov chains. The weights are assigned based on the concept of the R-measure, a theoretical framework inspired by universal coding principles, which serves as a consistent probability estimator for stationary and ergodic processes. Experimental results on real dataset are provided, demonstrating the potential of proposed method. This research underscores the potential of ensemble-based Markov models as a scalable and reliable tool for real-world forecasting tasks.

KW - Markov chains

KW - ensemble learning

KW - forecasting

KW - higher-order Markov models

KW - time series

UR - https://www.mendeley.com/catalogue/2efdcc30-1ba1-318b-9386-f39feb87a6e5/

UR - https://www.scopus.com/inward/record.uri?partnerID=HzOxMe3b&scp=105011288603&origin=inward

U2 - 10.1109/USBEREIT65494.2025.11054151

DO - 10.1109/USBEREIT65494.2025.11054151

M3 - Chapter

SN - 9798350392708

T3 - Proceedings - 2025 IEEE Ural-Siberian Conference on Biomedical Engineering, Radioelectronics and Information Technology, USBEREIT 2025

SP - 130

EP - 133

BT - Proceedings - 2025 IEEE Ural-Siberian Conference on Biomedical Engineering, Radioelectronics and Information Technology, USBEREIT 2025

PB - Institute of Electrical and Electronics Engineers Inc.

T2 - 2022 Ural-Siberian Conference on Biomedical Engineering, Radioelectronics and Information Technology (USBEREIT)

Y2 - 19 September 2022 through 21 September 2022

ER -