Tools
Tools
Patil, Abhay Udaykumar (2025) Machine Learning Model System for Multi-Crop Recommendation using Integrated Soil and Weather Insights for Western Maharashtra Region. Masters thesis, Dublin, National College of Ireland.
Preview |
PDF (Master of Science)
Download (1MB) | Preview |
Preview |
PDF (Configuration Manual)
Download (3MB) | Preview |
Abstract
The following thesis describes a machine learning-based, explainable multi-crop recommendation system designed in the Western Maharashtra region based on the combination of soil and weather data. Moreover, the paper combines a regional Kaggle dataset of 4,513 samples of macronutrients (N, P, K), soil pH, rainfall, temperature and categorical variables (district name and soil color) to train and compare three classifiers, Logistic Regression, Decision Tree, and Random Forest. With the help of structured preprocessing, the nutrient ratio and stratified validation features engineering, Random Forest with 100 trees and depth constraint is chosen as the final model due to its high Top-1 and Top-3 accuracy as well as the good generalisation across 16 crops. Leakage checks by label-shuffling, duplicate-row audits and input-noise sensitivity analysis are further used to measure model reliability with respect to ensuring that the patterns learned are agronomically meaningful and also do not change when realistic measurement uncertainty is present. SHAP is also incorporated to overcome the issue of opaqueness of ensemble models, giving global feature rankings and local explanations on each Top-3 recommendation of how individual soil and weather variables affect crop proposals. The resultant pipeline provides clear-cut Top-3 crop recommendations provided to the farmer based on unique soil testing at the western Maharashtra and other agro-climatic locations that can be integrated into digital decision-making tools to complement the advisory services.
Actions (login required)
 |
View Item |