A Hybrid Clustering Technique of SM-SOM for Detecting Abnormal Data of Listed Electrical Manufacturing Sector in P. R. China

For partitioning the dataset of financial ratios into abnormal and normal groups, this paper proposes a hybrid clustering technique by combining similarity matching (SM) algorithm with self-organizing maps (SOM), called SM-SOM technique. The hybrid system provides three stages: preprocessing stage, similarity matching with cosine algorithm and SOM cluster. For evaluating the performance of this hybrid technique, we give some experiments with quarterly financial ratios of listed electrical manufacturing sector in P. R. China. Here the financial ratios contain six categories: profitability, solvency, growth capability, risk level, cash-flow and operating ability, a total of 15 financial ratios are selected such as return on equity, net profit margin, liquidity ratio, and so on. The empirical results show that the SM-SOM technique can improve effectively the accuracy rate for clustering the financial data into normal and abnormal groups. Furthermore, using the hybrid technique we can find out which category these abnormal data fall into.