Similarity-based Time Series Prediction

• Grant number: 13/10978-8
• Support Opportunities: Scholarships in Brazil - Master
• Start date: September 01, 2013
• End date: August 31, 2015
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Gustavo Enrique de Almeida Prado Alves Batista
• Grantee: Antonio Rafael Sabino Parmezan
• Host Institution: Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil

Abstract

In the last decade we have seen a huge increase of interest for time series methods in Data Mining and Knowledge Discovery. Such interest has culminated in the proposal of literally hundreds of methods for tasks such as classification, cluste- ring, anomaly detection, motif detection, among others. Empirical research has demonstrated that time series similarity-based methods provide very competitive results, frequently outperforming more complex methods for tasks such as clas- sification, clustering and anomaly detection. We believe that the superiority of similarity-based methods is largely due to the community incessant work on dis- tance invariances such as warping, baseline, occlusion, rotation, etc. In contrast, statistical methods based on autoregression and moving averages are considered the state-of-the-art for time series modeling and prediction for over half-century now. In this research proposal, we ask if it really is the case, or if the Data Mining community has more to offer. In particular, we raise the hypothesis that similarity- based methods are simple, effective and parameter-light approaches for time series prediction. Although, similarity-based time series prediction methods have been re- searched in the recent past, we believe that previous research has failed to identify the correct invariances required for this task. The central hypothesis of this work is that only the right combination of offset, amplitude and recently-proposed com- plexity invariance, combined with a policy to avoid trivial matches leads to precise and meaningful predictions.

Scientific publications

(References retrieved automatically from Web of Science and SciELO through information on FAPESP grants and their corresponding numbers as mentioned in the publications by the authors)

SABINO PARMEZAN, ANTONIO RAFAEL; SOUZA, VINICIUS M. A.; BATISTA, GUSTAVO E. A. P. A.. Evaluation of statistical and machine learning models for time series prediction: Identifying the state-of-the-art and the best conditions for the use of each model. INFORMATION SCIENCES, v. 484, p. 302-337, MAY 2019. (18/05859-3, 16/04986-6, 13/10978-8)

PARMEZAN, ANTONIO R. S.; SOUZA, VINICIUS M. A.; BATISTA, GUSTAVO E. A. P. A.. Time Series Prediction via Similarity Search: Exploring Invariances, Distance Measures and Ensemble Functions. IEEE ACCESS, v. 10, p. 22-pg., 2022-01-01. (13/10978-8, 16/04986-6, 18/05859-3)

Academic Publications

(References retrieved automatically from State of São Paulo Research Institutions)

PARMEZAN, Antonio Rafael Sabino. Similarity-based time series prediction. 2016. Master's Dissertation - Universidade de São Paulo (USP). Instituto de Ciências Matemáticas e de Computação (ICMC/SB) São Carlos.