%0 Journal Article
%T What Can Artificial Intelligence Reliably Predict in Vector Biology? A Systematic Review of Analytical Tasks, Data Integration, Explainability, and Reproducibility
%A Michael Roberts
%A Sarah Thompson
%A James Anderson
%J Entomology and Applied Science Letters
%@ 2349-2864
%D 2026
%V 13
%N 1
%R 10.51847/ZXVYLu2H9N
%P 99-110
%X Artificial intelligence is increasingly used to classify vectors, automate surveillance, forecast abundance and transmission risk, and inform intervention planning. These applications may strengthen predictive entomology by integrating images, acoustic and optical signals, environmental observations, entomological records, and epidemiological data. However, the reliability of such predictions remains difficult to judge because analytical tasks, biological outcomes, validation designs, and intended decisions are frequently treated as comparable when they represent distinct constructs. This systematic review examines what artificial intelligence can defensibly predict in vector biology and under which biological, spatial, temporal, technical, and operational conditions those predictions remain credible. The review uses prespecified questions and eligibility boundaries, database-specific search logic, staged screening, structured evidence extraction, ecological adaptation of prediction-model bias and applicability appraisal, and narrative synthesis organized by analytical task and evidence class. It integrates evidence on vector identification, automated surveillance, abundance and distribution forecasting, transmission prediction, intervention-relevant classification, decision support, explainability, reproducibility, and external validity. The strongest synthesis is that artificial intelligence can provide useful task-specific predictions within bounded data-generating environments, particularly when targets, sensors, species, scales, and validation contexts are explicitly defined. Reliability weakens when internal performance is generalized to untested sites, devices, seasons, taxa, or decisions. Prediction accuracy does not by itself demonstrate decision usefulness; post-hoc explanation does not establish biological plausibility; and reproducible code does not ensure reproducible performance under distribution shift. Progress therefore depends less on adding model complexity than on independent validation, biologically coherent outcome definitions, leakage-resistant evaluation, transparent uncertainty, interoperable multimodal data, and prospective testing of whether predictions improve surveillance or control decisions.
%U https://easletters.com/article/what-can-artificial-intelligence-reliably-predict-in-vector-biology-a-systematic-review-of-analytic-lqxdmeacajsqbkg