< Back to previous page

Project

Multivariate time series forecasting and classification using intrinsically explainable machine learning models to solve real-world problems. (R-11811)

Time series data are ubiquitous. They are found in any domain which involves temporal measurement and are used for a variety of tasks such as health prognosis, financial forecasting, anomaly detection, human activity recognition, industrial maintenance, and others. Time series data can be univariate (single observation over time) or multivariate (multiple observations over time). The latter are under-represented in the literature due to their complexity. Two popular time series related tasks that have received extensive attention are (i) short-term and/or long-term forecasting and (ii) classification. The most efficient of the existing Machine Learning (ML) methods that solve such tasks operate in a black-box manner. Despite their relatively accurate results, this lack of transparency reduces their credibility in the eyes of end users. To tackle this issue, fuzzy logic has been employed. Fuzzy Cognitive Maps (FCMs) is an example of an interpretable graph-based knowledge representation method that has been successful to solve time series problems. This project emphasises on the application of hybrid learning FCM-based algorithms with strong mathematical foundations that are able to reason with human knowledge. Apart from FCMs, other approaches towards explainable Artificial Intelligence to be explored are hybrid models that combine ML with knowledge engineering and knowledge acquisition techniques or neuro-symbolic reasoning. Considering the above-mentioned points, this PhD project aims to employ inherently interpretable methods that allow human-machine interaction to solve multivariate time series forecasting and classification focusing on solving real-world problems. Additionally, a software product that allows the application of such methods on real-world datasets will be developed at the end of this project.
Date:1 Mar 2021 →  Today
Keywords:Algorithm, business informatics, Fuzzy Cognitive Maps
Disciplines:Data mining, Machine learning and decision making, Neural, evolutionary and fuzzy computation, Knowledge representation and machine learning