<![CDATA[Purpose – Early diagnosis of koi fish diseases remains constrained by limited expert availability and the dominance of static rule-based expert systems that are less adaptive to new cases. This study aims to develop a practitioner-informed Case-Based Reasoning (CBR) screening prototype for six koi fish disease categories using positive-evidence weighted symptom similarity. Design/methods/approach – This research used a Research and Development design involving knowledge acquisition from two koi cultivator practitioner-experts, representation of 15 clinical symptoms, and the Retrieve-Reuse-Revise-Retain cycle. The web-based system was implemented using CodeIgniter 4, PHP 8.0, and MySQL 8.0. Similarity used positive-evidence weighted Jaccard with an insufficient-evidence gate (Σsi < 2). Evaluation comprised a preliminary usability test with 15 respondents and robustness testing on 500 synthetic base profiles transformed under six disturbance scenarios. Findings – Under KB-v2 with positive-evidence similarity, the illustrative case ranked Cloudy Eye at 69.6% while Fin/Tail Rot scored 0.0% (no shared-absence inflation). On the pooled legacy set, overall accuracy was 65.0% (97.3% among scored cases), weighted F1-score was 0.76, 33.2% of cases returned insufficient evidence, and 5.4% of scored cases were ambiguous. The usability test yielded 57.7% Good, 37.7% Fair, and 4.4% Poor item responses. Research implications/limitations – The evaluation used synthetic data generated from the same practitioner knowledge matrix and has not been validated by aquatic veterinarians or real clinical field cases. Originality/value – The study contributes positive-evidence weighted similarity with an insufficient-evidence gate, a governed retain pathway for expert-confirmed cases, and robustness testing with decision-safety metrics under incomplete and noisy inputs.]]>
Copyrights © 2026
