Explainable weather forecasts through an LSTM-CBR twin system.

Pirie, Craig; Suresh, Malavika; Salimi, Pedram; Palihawadana, Chamath; Nanayakkara, Gayani

Abstract

In this paper, we explore two methods for explaining LSTM-based temperature forecasts using previous 14 day progressions of humidity and pressure. First, we propose and evaluate an LSTM-CBR twin system that generates nearest-neighbors that can be visualised as explanations. Second, we use feature attributions from Integrated Gradients to generate textual explanations that summarise the key progressions in the past 14 days that led to the predicted value.

Citation

PIRIE, C., SURESH, M., SALIMI, P., PALIHAWADANA, C. and NANAYAKKARA, G. 2022. Explainable weather forecasts through an LSTM-CBR twin system. In Reuss, P. and Schönborn, J. (eds.) Workshop proceedings of the 30th International conference on case-based reasoning (ICCBR-WS 2022), 12-15 September 2022, Nancy, France. CEUR workshop proceedings, 3389. Aachen: CEUR-WS [online], pages 256-260. Available from: https://ceur-ws.org/Vol-3389/ICCBR_2022_XCBR_Challenge_RGU.pdf

• Presentation Conference Type: Conference Paper (published)
• Conference Name: Workshops of the 30th International conference on case-based reasoning (ICCBR-WS 2022)
• Start Date: Sep 12, 2022
• End Date: Sep 15, 2022
• Acceptance Date: Jul 22, 2022
• Online Publication Date: May 11, 2023
• Publication Date: May 11, 2023
• Deposit Date: Jun 2, 2023
• Publicly Available Date: Jun 2, 2023
• Publisher: CEUR-WS
• Peer Reviewed: Peer Reviewed
• Pages: 256-260
• Series Title: CEUR workshop proceedings
• Series Number: 3389
• Series ISSN: 1613-0073
• Keywords: LSTM; XCBR; NLG; Integrated gradients; Forecasting; Visualisation
• Public URL: https://rgu-repository.worktribe.com/output/1977757
• Publisher URL: https://ceur-ws.org/Vol-3389/ICCBR_2022_XCBR_Challenge_RGU.pdf