Comparison of the Different Hierarchical Clustering Techniques for the Classification of Soils under Oil Palm in Nigeria

FULL TEXT in - | page 37-46

Abstract

Cluster Analysis has been very well used in soil classification, but sustainable guidelines are not readily available for the choice of appropriate clustering technique for soil data. This paper tested the robustness of five common agglomerative hierarchical clustering methods using soil data collected from 25 farmers’ field within the oil palm belt of Nigeria using the Cophenetic Correlation Coefficients (CPCC) while the Kappa coefficients of agreements were used to compare grouped formed by the hierarchical method and that of the taxonomic classification. From the dendrogram arising from the clustering analysis, shows some diversity of soils within the study locations. However, the groups formed by the five methods were composed of different numbers of soil individuals indicating that the different methods created different results. The result of the Cophenetic Correlation Coefficients (CPCC) shows the highest values for the Average linkage (0.7337065) while the Complete (0.7255978) ranked second with the Ward method (0.7102611) coming close while Centriod and the Single methods ranked 4th and 5th with a CPCC value of (0.6928591) and (0.5071803) respectively. Kappa coefficients of agreements were generally low for the five methods compared indicating that the classification by the traditional method and the hierarchical methods were not the same. The study therefore shows that the average linkage method ranked best in the classification of the soil data in the study location. From the study, it is suggested that researchers should evaluate carefully the methodology to apply before using any of the hierarchical clustering methods, and that simply applying a particular method to a data set and accepting the solution at face value will not be adequate.

Keywords: Cluster Analysis, Hierarchical clustering methods, Soil data, Oil palm belt,

References