Group Indexing of Fly Ashes Using Unsupervised Learning and Fuzzy Clustering Techniques

For long, fly ash is the most sought-after material in the construction industry. It is used as cementitious material as a supplement (SCM) to ordinary Portland cement (OPC) concrete as a replacement for cement. Recently, its usage as a geopolymer cement while developing geopolymer concrete (GPC) mixes has been well documented in bibliography. Apart from this, fly ash has found wide utility in every other realm of civil engineering as a material to reckon with. Fly ashes have been traditionally classified into two groups Group-C and Group-F as per ASTM. BIS is classified as Class F, Class C and Class P. These categorizations are hinged on chemical constituents and their contribution as a constituent material in concrete matrix. This paper presents the application of unsupervised machine learning techniques namely, k-means, k-medoids and fuzzy c-means algorithm have been applied to classify fly ashes on a data set of 400 instances. The chemical composition like SiO₂, Al₂O₃, Fe₂O₃, CaO and LOI have been considered as fly ash attributes. The data instances are clustered optimally into three groups. K-means and k-medoids algorithms were able to cluster with minimal overlapping with similar number of instances of each group. These optimal numbers of clusters were found using elbow method. However, fuzzy c-means algorithm grouped differently by attaching a fly ash data instance to an appropriate group with higher degree of belongingness of each data instance to a particular cluster. From the results obtained, the clustering rendered by fuzzy logic approach proved to be optimal as it placed the instances in groups affixing them with degree of belongingness. Based on this clustering, group characteristics are also elicited.