Ai-driven Price Prediction System For Direct Farmer-to- Market Consumer Agricultural Market

Abstract

Farmers Who Sell Directly To Consumers Through Digital Platforms Often Have No Idea What Price To Charge For Their Produce. They Rely On Heuristic Judgment And Informal Market Signals, Or Whatever Their Neighbor Charged Last Week. This Leaves Money On The Table During High-demand Periods And Causes Unnecessary Losses When The Market Is Already Saturated. On The Other Hand, Consumers Have No Way To Tell Whether A Listed Price Is Fair Or Inflated. This Paper Tackles That Specific Problem By Building A Price Prediction System Called AIPPS (Agricultural Intelligent Price Prediction System) That Is Designed From The Ground Up For Farmer-to-consumer (F2C) Digital Markets, Not The Wholesale Exchanges That Most Existing Research Focuses On. AIPPS Combines A Bidirectional LSTM Network With An XGBoost Ensemble In A Two-stage Architecture. The Bi-LSTM Handles The Sequential Price History And Weather Patterns, While XGBoost Cleans Up Residual Errors By Incorporating Structured Features Like Supply Volumes, Transport Costs, And Seasonal Indicators. We Trained And Tested The Model On 24 Months Of Transaction Data From 120 Micro Markets Covering Five Commodity Groups. The System Achieves A MAPE Of 1.84%, RMSE Of 1.84, And R² Of 0.963, Which Is Substantially Better Than ARIMA, SVR, Random Forest, And Standalone LSTM Baselines. We Also Built A Mobile Dashboard For Farmers Showing A Simple Sell/hold Recommendation And A Consumer-facing Page That Shows Predicted Fair Price Ranges. Ablation Experiments Confirm That Each Component Of The Architecture Genuinely Contributes To The Final Accuracy.

Keywords