Jomenz/e-lightgbm-soil-fertility-ghana

An Engineered LightGBM-SHAP Model for Soil Fertility Classification and Fertiliser Recommendation — KNUST CS Research Proposal 2026 # E-LightGBM: Soil Fertility Classification & Fertiliser Recommendation **Author:** Odai Joel Nii Afotey **Student ID:** 21019091 **Institution:** Department of Computer Science, KNUST, Ghana **Supervisor:** Dr. Eric Osei **Year:** 2026 --- ## Project Overview This repository contains the full source code for the research proposal: > *"An Engineered LightGBM-SHAP Model for Soil Fertility Classification > and Fertiliser Recommendation: A Public Dataset Study on Ghanaian > Smallholder Farms"* The study develops **E-LightGBM** — an engineered variant of LightGBM that integrates a SHAP TreeExplainer module and a rule-based fertiliser recommendation layer — to classify soil fertility levels (Low / Medium / High) and generate farmer-interpretable fertiliser recommendations. --- ## Engineering Contribution - **Operation:** [F] FABRICATE - **What was added:** SHAP TreeExplainer module + fertiliser recommendation output layer - **Engineered model name:** E-LightGBM --- ## Baseline Models | Model | Role | |-------|------| | Standard LightGBM | Scientific control (unengineered) | | XGBoost | Gradient boosting competitor | | Random Forest | Non-boosting ensemble benchmark | ---