cff-version: 1.2.0
message: "If you use this software or its companion paper, please cite as below."
title: "agri-analyze: when do hand-crafted rules help modern CNNs for agricultural disease classification?"
abstract: >
  Research codebase for a multi-architecture evaluation of hand-crafted rule
  engines, learned decision-tree rules, and LLM-generated rule baselines
  composed with modern CNN classifiers on agricultural disease datasets.
  Includes reproducibility tooling, statistical protocol (bootstrap,
  Holm-Bonferroni, Dietterich 5x2cv, Friedman-Nemenyi), expected
  misclassification loss (EML) sensitivity analysis, and a mega-dataset
  training pipeline combining 14 public Roboflow/Kaggle rice+wheat sources
  (~30 K de-duplicated images, 22 canonical classes).
type: software
version: "0.3.0-mega-dataset"
license: MIT
url: "https://github.com/Ashut0sh-mishra/agri-analyze"
repository-code: "https://github.com/Ashut0sh-mishra/agri-analyze"
authors:
  - family-names: "Mishra"
    given-names: "Ashutosh"
    affiliation: "Independent Researcher, India"
    email: "mishra.ashutosh@gmail.com"
    orcid: "https://orcid.org/0009-0000-4764-8160"
keywords:
  - agricultural-ai
  - plant-disease-classification
  - rule-based-ai
  - neuro-symbolic
  - reproducibility
  - benchmarking
  - mega-dataset
preferred-citation:
  type: article
  title: "When Do Hand-Crafted Rules Help Modern CNNs for Agricultural Disease Classification?"
  authors:
    - family-names: "Mishra"
      given-names: "Ashutosh"
  year: 2025
  journal: "Smart Agricultural Technology (under review)"