Agrongo: A Machine Learning–enabled Platform For Farm-to-consumer Commerce And Sustainable Food Redistribution

Abstract

A Major Challenge That The Agriculture Supply Chain Is Facing In India Is The Lack Of Technology. Fragmented With Large Losses In Food Post Harvest, Lowering Farmers' Income, Poor Food Availability For Low-income Groups. This Paper Outlines A Web Based Platform That Creates A Direct Market Between Farmers, Thereby Not Only Catering To The Needs Of Consumers But Also Enabling The Distribution Of Near-expiry Produce In Agriculture To Registered Non Governmental Organizations (NGOs). They Combine A System That Integrates A Random Forest Ensemble Classifier For Analysing Uploaded Images To Predict The Freshness Score Of Fruits And Vegetables Using OFV Estimated Shelf Life. If The Forecasted Expiry Period Falls Below A Certain Threshold (default: 2 Days), The System Automatically Sends Email Alerts To Farmers And All Registered NGOs, So As To Sell/donate Them On Time Before It Becomes A Waste Produce. The Platform Is Implemented Developing The Front-end Using HTML, CSS, And JavaScript And The Backend Using PHP; MySQL As The Relational Database; And Python For The Machine Learning Module. Experimental Evaluation On A Labelled Data Set Of 4200 Fresh And Near-expiry Produce Images Across 12 Different Categories Of Common Indian Agricultural Commodities Demonstrates That The Random Forest Classifier Has An Accuracy Of 93.2%, A Precision Of 92.7%, A Recall Of 0.934, And An F1-score Of 0.932. A 45-day Pilot Deployment Resulted In A 27% Rise In Farmer Profitability, A 19% Reduction In Consumer Prices, The Prevention Of 340 Kg Of Food Wastage, And An Estimated 2,800 Meals Served To NGO Beneficiaries.

Keywords