{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,2,23]],"date-time":"2026-02-23T10:58:41Z","timestamp":1771844321827,"version":"3.50.1"},"reference-count":0,"publisher":"Slovenian Association Informatika","issue":"8","content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["IJCAI"],"abstract":"<jats:p>Early detection of plant diseases is crucial for global food security. While Deep Learning ensemble techniques are widely adopted to improve performance, the assumption that simply aggregating models is always beneficial should be nuanced. This paper addresses this issue by conducting a rigorous analysis of an ensemble of four state-of-the-art architectures (Swin Transformer, Vision Transformer, EfficientNetV2, ConvNeXt) on the PlantDoc dataset, a benchmark known for its complexity.Our approach is twofold. First, we conduct a systematic ablation study to assess how each model contributes to the ensemble's performance. This analysis leads to the counter-intuitive finding that an optimized three-model subset (Swin, ViT, and EfficientNetV2) outperforms the full four-model ensemble. Quantitatively, the pruned ensemble achieves a Macro F1-score of 0.7503 and an accuracy of 0.7619, compared to 0.7409 and 0.7500 for the full set, respectively. Second, to explain this phenomenon, we perform a prediction correlation analysis. This reveals significant predictive redundancy, stemming from the architectural similarities between ConvNeXt and the Transformer-based models, as the cause for this sub-optimality. These findings suggest a key principle for ensemble design: the predictive complementarity of the models is a more critical factor than their individual performance or the complexity of the aggregation method-a finding reinforced by our benchmark showing that even advanced strategies like calibrated voting and stacking failed to outperform the pruned ensemble. Our work thus positions the methodological pairing of ablation study and correlation analysis as an essential and pragmatic approach to optimize the performance of ensembles in computer vision.<\/jats:p>","DOI":"10.31449\/inf.v50i8.10277","type":"journal-article","created":{"date-parts":[[2026,2,22]],"date-time":"2026-02-22T11:59:36Z","timestamp":1771761576000},"source":"Crossref","is-referenced-by-count":0,"title":["Optimizing Deep Learning Model Ensembles for Plant Disease Detection through Ablation and Correlation Analysis"],"prefix":"10.31449","volume":"50","author":[{"given":"Nawel","family":"Ghrieb","sequence":"first","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Tahar","family":"Guerram","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]},{"given":"Othaila","family":"Chergui","sequence":"additional","affiliation":[],"role":[{"role":"author","vocabulary":"crossref"}]}],"member":"16141","published-online":{"date-parts":[[2026,2,21]]},"container-title":["Informatica"],"original-title":[],"link":[{"URL":"https:\/\/www.informatica.si\/index.php\/informatica\/article\/download\/10277\/6526","content-type":"application\/pdf","content-version":"vor","intended-application":"text-mining"},{"URL":"https:\/\/www.informatica.si\/index.php\/informatica\/article\/download\/10277\/6526","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,2,23]],"date-time":"2026-02-23T10:01:37Z","timestamp":1771840897000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.informatica.si\/index.php\/informatica\/article\/view\/10277"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,2,21]]},"references-count":0,"journal-issue":{"issue":"8","published-online":{"date-parts":[[2026,2,21]]}},"URL":"https:\/\/doi.org\/10.31449\/inf.v50i8.10277","relation":{},"ISSN":["1854-3871","0350-5596"],"issn-type":[{"value":"1854-3871","type":"electronic"},{"value":"0350-5596","type":"print"}],"subject":[],"published":{"date-parts":[[2026,2,21]]}}}