Smart Crop Recommendation System

link to the video: https://youtu.be/nVsl1SV6peM?si=QAl62YudmvHhIrHM

Overview

This project implements a Crop Recommendation System that suggests the most suitable crop to grow based on soil nutrients and environmental conditions. It combines Machine Learning (ML) with IoT (ESP8266 microcontroller) to support precision agriculture. The workflow: Collected and preprocessed agricultural data (soil nutrients & climate). Trained a Random Forest Classifier model to recommend crops. Built a Flask web application for user-friendly crop prediction and farming advice. Converted the trained ML model to C code (micromlgen) and deployed it on an ESP8266 NodeMCU microcontroller.

Dataset The dataset includes soil and environmental features: N: Nitrogen content ratio in soil P: Phosphorus content ratio in soil K: Potassium content ratio in soil Temperature: in °C Humidity: in % pH: Soil acidity/alkalinity Rainfall: in mm

Tech Stack

Machine Learning: Random Forest Classifier (scikit-learn)

Web Framework: Ruby and Flask

Frontend: HTML, CSS (simple UI for predictions)

Microcontroller: ESP8266 (NodeMCU)

TinyML Library: micromlgen for C model export

Flask Web App The web app allows users to: Input soil & environmental parameters. Get the best crop recommendation. Receive advice on optimizing conditions (nutrients, pH, rainfall, etc.).

Ruby Backend With ONNX Runtime The second deployment path validates language-agnostic inference by converting the trained Random Forest model to ONNX format.

Key Steps

Export trained model from Python → ONNX

Load ONNX model in Ruby using onnxruntime gem

Run inference directly from Ruby backend code

Serve predictions to a lightweight Ruby-based web endpoint or CLI script

Purpose

Demonstrates cross-platform, cross-language model deployment

Makes the engine compatible with Ruby web stacks (e.g., Rails, Sinatra)

Confirms interoperability without retraining or rewriting ML code

Ruby Deployment Capabilities

The Ruby ONNX version can:

Accept soil parameters via a web form or API

Run fast inference locally using ONNX Runtime

Return the crop recommendation + condition improvement notes

ESP8266 Deployment The trained Random Forest model was converted into C code. Deployed on ESP8266 NodeMCU with: DHT11 sensor for temperature & humidity. LCD display to show predicted crop. Random values (for testing) simulating N, P, K, pH, rainfall. Feel free to fork, improve, or suggest enhancements. Together, let’s make farming smarter