Comparative Analysis of Feature Importance and Correlation of Soil Components in Maize Cultivation Using Machine Learning

Publication Type

Journal Article

Publication Date (Issue Year)

2026

Journal Name

Data Science and Applications

Abstract

This study analyses the relationship between soil components and maize yield using three Machine Learning (ML) models: K-Nearest Neighbours Regressor (KNNR), Gradient Boosting Regressor (GBR), and Random Forest Regressor (RFR). Twelve soil properties: pH, nitrogen (N), organic carbon (OC), phosphorus (P), calcium (Ca), magnesium (Mg), potassium (K), sodium (Na), zinc (Zn), copper (Cu), manganese (Mn), and iron (Fe) were analysed for their predictive relevance. Feature importance analysis and correlation coefficients revealed that iron (0.3436) and manganese (0.3022) had strong positive associations with maize yield, while pH (−0.0727), organic carbon (−0.0692), and phosphorus (−0.0589) exhibited negative relationships. This emphasises the critical role of micronutrient availability and soil acidity in influencing crop performance. Among the models, RFR demonstrated superior predictive accuracy with the lowest error metrics (MAD: 1.07; MSE: 2.13; RMSE: 1.46). Paired t-tests confirmed that RFR significantly outperformed KNNR (p = 0.0109), while its difference from GBR was not statistically significant (p = 0.4953). These findings highlight RFR as a reliable model for yield forecasting and emphasize the value of detailed soil component analysis in guiding precision nutrient management for enhanced maize productivity.

Keywords

Comparative Analysis, Feature Importance, Correlation, Soil Components, Maize Cultivation, Machine Learning

Rsif Scholar Name

Daniel Dzarma Ezra

Rsif Scholar Nationality

Nigeria

Cohort

Cohort 4

Thematic Area

ICTs Including Big Data and Artificial Intelligence

Africa Host University (AHU)

Université d'Abomey-Calavi, Benin

Link to Full Text

Share

COinS