Abstract:
Food insecurity on a global scale still affects millions of individuals, necessitating
researchers to identify and address the underlying factors. Despite numerous efforts
to mitigate food insecurity, significant gaps remain in understanding the effects of
socioeconomic and climatic factors on major crop production, such as maize which
affects food insecurity. Numerous statistical and machine learning approaches have
been utilized in addressing the problem, however, some of these approaches cannot
accurately and robustly model the underlying structures of the data. For example,
while machine learning approaches may produce accurate predictions, they are less
interpretable than traditional statistical models, thus a need to identify alternative
more interpretable models that produce accurate and robust predictions. Therefore,
the study aims to compare the performance of the K-Means algorithm and the Gaussian
Mixture Model (GMM) for clustering, as well as the K-Nearest Neighbor (KNN)
and Random Forest (RF) algorithms for classifying global agricultural production of
maize. The K-Means and Gaussian Mixture Model (GMM) cluster countries based on
maize production and food insecurity, evaluated using metrics like the Silhouette Coefficient, Dunn’s Index, and Davies-Bouldin Index, while K-Nearest Neighbor (KNN)
and Random Forest (RF) classify production categories, assessed with accuracy, precision, recall, F1-score, ROC, and AUC. Features such as agricultural land, food
insecurity level, population, and climatic factors including CO2 emissions, temperature,
and precipitation were collected from multiple online datasets and databases
for the year 2022. The findings of the study indicate that the K-means outperformed
the GMM algorithm and the RF produced better results than the KNN algorithm in
predicting maize production categories. Furthermore, from the two distinct country
clusters, cluster one has higher maize yields and lower food insecurity while cluster
two has lower maize yields and higher food insecurity levels. These results might
assist policy interventions on mitigating climate impacts and also suggest sustainable
agricultural practices in high-risk regions, like Sub-Saharan Africa, Southern Asia,
and Central America.
