AgriTrend-DSS: An Agricultural Decision Support System Integrating Machine Learning and Generative AI

Core Overview AgriTrend-DSS is an open-source agricultural decision support system designed for local farmers, combining machine learning (ML) for yield prediction and generative AI for market trend analysis to enable data-driven planting decisions.

Key Basics

• Author/Maintainer: riyandimuhamad
• Source: GitHub (https://github.com/riyandimuhamad/AgriTrend-DSS)
• Release Date: 2026-06-04

This system addresses small farmers' information asymmetry and unreliable traditional decision-making by providing scientific tools for yield forecasting and market insights.

Small farmers often face:

• Lack of scientific yield prediction tools
• Difficulty in grasping market dynamics
• Unreliable decisions based on experience/intuition (exacerbated by climate change and market volatility)

AgriTrend-DSS was built as an open-source solution to these issues. It merges ML and generative AI to offer data-driven support, helping farmers decide what to plant and when to sell.

1. Yield Prediction Engine: Uses Random Forest algorithm on agricultural data (NPK content, temperature, humidity, pH, rainfall) to predict yield per hectare.
2. Market Trend Classification:
   • Uptrend Potential: ≥6.5 tons/ha → expand planting
   • Stable Market:4.0-6.4 tons/ha → maintain strategy
   • Adjust Distribution: <4.0 tons/ha → find alternative markets or crops
3. AI Insights: Integrates Google Gemini API (gemini-2.5-flash) with two modes:
   • Market-only analysis (trends, pricing)
   • Agronomic + market analysis (tech advice + market insights)
4. Geo Coordinate Parsing: Built-in Indonesia region database to resolve lat/long via region code/name for regional analysis.

Frontend: React19, TanStack Router, Vite, TypeScript, Tailwind CSS Backend: FastAPI, Python3.12, Uvicorn Data & Auth: Supabase (PostgreSQL + auth), JWT ML & AI: scikit-learn (Random Forest), Google Gemini API, joblib (model serialization) Deployment: Vercel (backend API), Cloudflare (frontend CDN)

1. Data Collection: Input soil (NPK), weather (temp, humidity, rainfall), pH data.
2. Crop Selection: Choose crop type (rice, corn etc.).
3. Region Info: Enter region code/name for geo parsing.
4. Analysis Mode: Select market-only or agronomic+market analysis.
5. Smart Prediction: ML computes yield → trend classification → AI generates structured report (insights, 30-day actions, tech improvements).
6. Decision Support: Get yield forecast, trend judgment, and actionable advice.

For Small Farmers:

• Access enterprise-level decision tools
• Evaluate crop收益 before planting
• Adjust sales timing based on trends
• Get targeted agronomic advice

For Cooperatives:

• Unified decision reference for members
• Regional data analysis
• Coordinate production to avoid oversupply
• Improve bargaining power

For Policymakers:

• Understand regional production potential
• Design targeted agricultural policies
• Predict food security risks
• Optimize resource allocation

Open Source:

• License: ISC (free use, modify, distribute; commercial allowed)
• Modular architecture: Easy to add new ML models, regional data, AI services, or mobile apps.

Future:

• With more data and model optimization, it will play a bigger role in global food security and sustainable agriculture.
• Invites developers/researchers to contribute to this open-source project.