Surprise-Driven Learning: Automated Detection and Analysis of Anomalous Prototype Retrievals to Drive System Evolution

The heuristic of “investigating deviations from expectations” is rarely automated in CBDS. While Koornneef et al. (2021) and Shved et al. (2024) focus on retrieval efficiency and accuracy, they don’t address what happens when the system’s recommendations surprise or disappoint users. This idea proposes embedding anomaly detection algorithms (potentially leveraging unsupervised ML or user feedback monitoring) to continuously identify retrievals that produce unexpected outcomes—such as prolonged decision times, low user acceptance, or post-hoc corrections. These flagged instances would trigger deeper analysis: are there missing features, outdated cases, or systemic biases? The system would then suggest updates to the case library or retrieval algorithms, creating a feedback loop of continual improvement. This “surprise-driven” evolution could make CBDS systems more robust, adaptive, and aligned with real-world needs, advancing beyond static or purely accuracy-focused methods.

References:

1. A Web-Based Decision Support System for Aircraft Dispatch and Maintenance. H. Koornneef, W. Verhagen, R. Curran (2021). Aerospace.
2. INTELLECTUAL SUPPORT OF THE PROCESSES OF SEARCHING AND EXTRACTION OF PRECEDENTS IN CASE-BASED REASONING APPROACH. A. Shved, Ye. O. Davydenko, H. V. Horban (2024). Radio Electronics, Computer Science, Control.