A HYBRID INTERNET OF THINGS AND MACHINE LEARNING APPROACH FOR A USER-FRIENDLY PRECISION AGRICULTURE FRAMEWORK.

Abstract

Objectives: To develop an IoT and Machine Learning (ML) based decision-making framework that enables farmers to make informed decisions through data-driven approaches on the suitable type of crop to be planted, corresponding to respective soil conditions. Additionally, a Nutrient management approach to maintain soil health and optimize healthy yields.

Methods: Two primary models have been implemented: a Crop Recommendation model that suggests crops suitable for the respective soil type by analysing soil health using real-time crucial parameters such as nitrogen (N), phosphorus (P), potassium (K), pH, rainfall, temperature, and humidity obtained via a NPK sensor, and a Nutrient Sufficiency Classification model that analyses the sensed nutrient content and classifies nutrient sufficiency (low, sufficient, high) for N, P, and K. The output of the Nutrient Sufficiency Classification model is linked to a rule-based feedback system that provides suggestions or solutions for the detected shortcomings to improve soil health.

Findings: The Crop Recommendation model demonstrated an accuracy of 90.35% and the Nutrient Sufficiency Classification model exhibited an accuracy of 85.12%. Soil parameters analysed by the model, which include: nitrogen (N), phosphorus (P), potassium (K), pH, Rainfall, Temperature, and humidity, were either manually given as input through the web-based interface or a combination of real-time sensing and third-party APIs, including Google and weather APIs, that auto-populated the input fields in the user-friendly HTML-based web interface. The entire system was validated for its performance. The predictive results of the ML model were displayed on the web interface.

Novelty: This study implements a practical and deployable framework with a real-time sensing IoT architecture to sense nutrient content in the soil, an ML framework for data analysis with an integrated rule-based feedback engine to auto-populate the web interface that gives crop recommendation outputs and actionable feedback in case of nutrient deficiencies.