CODA - Predictive Models with Interpretability and COncept Drift Analytics

Important results from the project

The purpose was to develop more knowledge about data-driven predictive models of vehicles with focus on concept drift and interpretability. The project has contributed to strengthening the above research area by developing methods and knowledge for analysis of concept drift and dynamic data and analysis and improvement of interpretability. The project has also developed a framework/architecture for deploying ML-models, a process that can result in better requirements for data collection and processes/guidelines for how external users should have access to a company´s data.

Expected long term effects

The project delivered two doctoral thesis and a number of articles and degree projects, which give other researchers and companies the opportunity to take part of the results in predictive modeling and further develop the methods and knowledge. Knowledge from the project have been used in the research project FAMOUS. The project also developed software for interpretability, framework/architecture for deployment of ML models, a demonstrator for data collection and processes/guidelines for data sharing with external users. These results are used in Scania´s R&D.

Approach and implementation

The participants were Scania CV, Linköping University, Stockholm University and Royal Institute of Technology. The project comprised six work packages: Project management, Data drift and dynamic data, Interpretability of models, Deployment of prognostic models, Configurable data collection and anomaly detection and Data warehouse. 13 workshops were held with the project group, as well as several pulse meetings. Each workshop began with a management meeting and then results from each work package were presented. Two extra workshops were held at the end, one of which was public.