ML Based Soil Health Assessment and Fertilizer Recommendation System Using IoT

Abstract: Soil health is a fundamental determinant of agricultural productivity, yet conventional testing methods remain costly and incapable of real-time feedback. This paper presents AgriSmart - a physically implemented IoT-based Soil Health Assessment and Fertilizer Recommendation System. An ESP32 Wi-Fi microcontroller (Device ID: ESP32-AGRISMART-001) is interfaced with a capacitive soil moisture sensor and a DHT11 temperature-humidity module. Soil pH is determined using the distilled-water pH paper method and entered manually via the web dashboard. Rainfall data is fetched in real time using the OpenWeatherMap API. All sensor readings are transmitted via HTTP POST in JSON format to a Django backend, stored in an SQLite database, and processed by a three-model Random Forest pipeline: Soil Type classification (71.8%), Soil Health assessment (88.2%), and Fertilizer Recommendation (93.6%). The system was validated with real soil samples and supports 12 fertilizer classes across 6 soil types and 22 crop varieties. Results confirm practical viability for precision agriculture.

Keywords: IoT, ESP32, AgriSmart, Soil Health, Fertilizer Recommendation, Random Forest, Django, SQLite, DHT11, OpenWeatherMap API, Precision Agriculture, Soil Type Classification