SOIL FERTILITY GRADE PREDICTION USING MACRO AND MICRONUTRIENT COMPOSITION ANALYSIS

Abstract

Soil fertility is a fundamental determinant of agricultural productivity, directly influencing crop
yield, food security, and sustainable land management. Conventional approaches to soil fertility
assessment are time-consuming, laboratory-intensive, and often inaccessible to smallholder
farmers in developing regions. This study proposes a machine learning-based framework for the
automated prediction of soil fertility grades by analysing the composition of both macro and
micronutrients present in soil samples.
The macronutrients considered include Nitrogen (N), Phosphorus (P), and Potassium (K), while
the micronutrients encompass Iron (Fe), Zinc (Zn), Copper (Cu), Manganese (Mn), and Boron
(B), along with supplementary parameters such as soil pH, organic carbon content, and electrical
conductivity. A curated dataset of labelled soil samples was subjected to rigorous preprocessing,
including outlier removal, normalisation, and feature selection using correlation analysis and
feature importance ranking.
Multiple supervised classification algorithms — including Random Forest, Gradient Boosting,
Support Vector Machine (SVM), and K-Nearest Neighbours (KNN) — were trained and
evaluated to predict discrete fertility grade categories (High, Medium, Low, and Deficient). The
Random Forest classifier achieved the highest predictive accuracy of 94.3%, demonstrating
superior performance in handling multi-class imbalance and non-linear nutrient interactions.
Experimental results validated using k-fold cross-validation confirm the robustness and
generalisation capability of the proposed model