hodiembo/dot-vs-full-refraction-recommender-system

Machine learning based recommendation system for classifying participants into DOT Glasses or Full Refraction using screening, prescription, demographic, and process-related data from low-resource eye-care settings in Kenya. # dot-vs-full-refraction-recommender-system Machine learning based recommendation system for classifying participants into DOT Glasses or Full Refraction using screening, prescription, demographic, and process-related data from low-resource eye-care settings in Kenya. This repository contains the implementation notebook for a machine learning based recommendation system that classifies participants into either the DOT Glasses pathway or the Full Refraction pathway. ## Project Title Development of a Data-Driven Recommendation System for Eyewear Selection in Low-Resource Settings ## Project Overview The project was developed to support decision making in low-resource eye-care settings where community health workers often need to decide whether a participant should receive simplified DOT Glasses or be referred for full refraction. The system uses participant screening, prescription, demographic, and process-related variables to generate a recommendation. ## Objective To develop a data-driven recommendation system that predicts the most suitable eye-care pathway between simplified DOT Glasses and full refraction. ## Methods The implementation involved: - data loading and merging - preprocessing and logMAR standardization - feature preparation - model training and comparison - validation and evaluation - deployment packaging The candidate models tested were: - Logistic Regression - Decision Tree - Random Forest - Gradient Boosting ## Final Model Logistic Regression was selected as the final model because it gave the strongest and most stable performance across the experimental stages. ## Final Performance - Accuracy: 64.6% - Precision: 64.2% - Recall: 55.8% - F1-score: 59.7% - ROC-AUC: 62.5% ## Repository Contents - `usiu_project_implementation.ipynb` — main implementation notebook - additional outputs and figures where applicable ## Dataset The dataset used in this project is stored separately and can be accessed through the project dataset link provided in th …