<i>K</i>-means cluster analysis of the West African species of cereals based on nutritional value composition

Abstract

The K-means algorithm was deployed to extract clusters within the prevalent cereal foods in West Africa. The West Africa Food Composition Table (WAFCT) presents all the 76 food sources in the cereals class as a single group without considering the similarity or dissimilarity in nutritional values. Using K-means clustering, the Euclidean distance between nutritional values of all cereal food items were measured to generate six sub- groups based on similarity. A one-way analysis to validate the results of the extracted clusters was carried out using the mean square values. For every nutrient, the “within groups” and “between groups” values of the mean squares were examined. This was done to ascertain how similar or dissimilar data points in the same or different clusters were to each other. It was discovered that the P values for all “between groups” and “within groups” mean squares for every nutrient was P < 0.01. Additionally, it was observed that in all cases, the mean square values of the “within groups” were significantly lower than those of the “between groups”. These outcomes are indications that clustering was properly done such that the variability in nutrient values for all food sources within the same clusters was significantly low, while those in different clusters were significantly
high. Thus, the ultimate objective of clustering, which is to maximize intra-cluster similarity and minimize inter-cluster similarity was effectively achieved. Cluster analysis in this study showed that all food items within a particular cluster are similar to each other and dissimilar to food items in a different cluster. These findings are valuable in dietaries, food labeling, raw materials selection, public health nutrition, and food science research, when answering questions on the choice of alternative food items. Where original choices are not available or unaffordable, the clusters can be explored to select other similar options within the same cluster as the original choice.