An Application of Unsupervised Clustering Technique to Finance Data

This paper deals with an application of unsupervised clustering technique in finance using the data pertaining to top ranking private sector companies in India for the three randomly selected financial years. A clustering algorithm which makes use of statistical distance measures and self-organizing principles of Artificial Neural Network (ANN) is used in this paper to group the companies which are rated as topmost companies in terms of gross sales. Only profitability ratios are considered for the study as the other ratios did not provide useful information in the formation of clusters. The clustering technique applied here makes use of centers obtained from K-means algorithm, without iteration, as initial weights. These weights connect the input nodes to the nodes in the output layer. The clustering technique updates the weights of the winning nodes and the neighbourhood nodes. The final clusters are obtained using the clustering method. The clustered data set is then subjected to discriminant analysis repeatedly by considering the classification of groups obtained in iteration ‘t’ as input into the next iteration ‘t+1’, until a cent percent classification is achieved. The main purpose of the study is to categorize the companies into different grades in terms of the ratios. The cluster analysis yielded three groups for each of the three periods. The clusters are arranged in the order of their centroids and the companies are graded as A, B or C. Grade A companies have low-profile in terms of the ratios while, Grade B and C are companies with medium-profile and highprofile ratios.