Reproducibility data and analysis for "Quantifying Training-Seed and Data-Partition Variability in Deep Learning for Plant Disease Classification"

This repository contains the formal experimental outputs and analysis files associated with the study “Quantifying Training-Seed and Data-Partition Variability in Deep Learning for Plant Disease Classification.” Three ImageNet-pretrained architectures, MobileNetV2, EfficientNetB0 and ResNet50, were evaluated in two controlled experiments comprising 60 formal runs in total. Experiment A used one fixed stratified data partition and ten training seeds, whereas Experiment B used one fixed training seed and ten stratified data partitions. The repository contains run-level metrics, split manifests, per-image predictions, training histories, protocol and computational-environment metadata, derived statistical tables, manuscript-level analysis outputs and the formal analysis notebook. The original Cassava Leaf Disease Classification images are not redistributed and should be obtained from their original Kaggle source.