Reproducibility and mobile deployment benchmark data for lightweight CNN banana leaf disease classification
Description
This dataset provides reusable reproducibility and mobile deployment benchmark data generated from a comparative study of EfficientNetV2B0, MobileNetV3S, and SqueezeNet1_1 for three-class banana leaf disease classification. The package includes the final processed-dataset partition manifest, laboratory and field corruption-robustness workbooks and machine-readable derivatives, McNemar verification results, Android benchmark photographs and aggregate latency/RAM measurements, CRITIC–TOPSIS decision data, sensitivity-analysis records, software configuration, provenance documentation, limitations, and integrity checksums. Original source images, the processed image collection, and the external out-of-distribution dataset remain hosted in their respective public repositories and are linked rather than redistributed.