Dataset, Model and Notebooks Repository Finger Millet Dataset

This repository accompanies the project "A Machine Learning Model for Predicting Finger Millet Grain Weight to Support Agronomic Decision Making in Uganda." It contains the curated dataset, trained model artefacts, and Jupyter notebooks used to develop and evaluate a predictive model of finger millet (Eleusine coracana) grain weight from agronomic and environmental variables. Background. Finger millet is a climate-resilient staple and income crop for smallholder farmers across eastern and northern Uganda, yet yield prediction remains largely experience-based. Grain weight is a key yield component, and being able to estimate it before harvest from measurable agronomic traits gives extension officers, researchers, and farmers an evidence base for decisions on variety selection, spacing, fertiliser application, and harvest planning. Contents. data/ - the finger millet dataset ([N] records, [M] variables) collected from [trial sites / seasons / source], covering variables such as [plant height, tillers per plant, panicle length, number of fingers, days to maturity, soil parameters, rainfall, variety]. Includes raw and cleaned versions plus a data dictionary. models/ - serialised trained models ([e.g. Random Forest, XGBoost, Linear Regression]) with associated preprocessing pipelines and hyperparameter settings. notebooks/ - reproducible Jupyter notebooks covering exploratory data analysis, preprocessing and feature engineering, model training and comparison, evaluation, and feature-importance analysis. results/  evaluation metrics ([RMSE, MAE, R²]) and figures. Methods. Records were cleaned and screened for outliers and missing values, split into training and test sets ([e.g. 80/20]) and evaluated with [k-fold cross-validation]. Model performance was compared across [algorithms], with the best-performing model achieving [R² = X, RMSE = Y g]. Intended use. The resources support reproducibility of the reported results and reuse by researchers working on crop yield prediction, agricultural data science, and decision-support tools for smallholder systems in sub-Saharan Africa. The dataset may also serve as a teaching resource for applied machine learning in agriculture. Limitations. The data reflect the agro-ecological conditions of [study region] over [seasons]; predictions should be validated locally before use in other zones. Funding/affiliation. Developed at Makerere University, college of computing and information sciences, Department of Information Technology , as part of Masters of Science in Information Technology.  Citation. Maloomo, K. M. ([2026]). Dataset, Model and Notebooks Repository: Finger Millet Dataset. Zenodo. Dataset, Model and Notebook…