A Kdd Algorithm And Its Application

In the past thirty years, the steady, startling development of computer hardware has resulted in a large supply of powerful computers, data-collecting devices and storage media. These techniques have greatly accelerated the advancement of database and information industry, which produced a great deal of data and information conducing to transaction management, information retrieval and data analysis. Because of great data quantity, it's far beyond the ability of human to understand the knowledge hidden in data, which is the phenomenon known as rich in data but short in knowledge. Thus database management system for data storage, in combination with machine learning for data analysis to mine the knowledge behind great amount of data, gave birth to Knowledge Discovery in Database (KDD). In fact, KDD is a cross-subject relating to machine learning, pattern recognition, statistics, intelligent database, knowledge acquisition, data visualization, high performance computing and expert system, etc. It's the nontrivial Process of identifying valid, novel, potentially useful, and ultimately understandable pattern in data. By KDD, we can apply the knowledge to practical data processing, supporting scientific decision. The phase of knowledge learning in KDD is called Data Mining (DM), whose algorithm is core of KDD system. This paper is based on process of knowledge learning in KDD, emphasizes particularly on DM algorithm, especially on artificial neural networks such as Self-Organized feature Map, and is combined with mathematical statistics. We put forward a feasible KDD clustering algorithm, DASOM, implemented in analogous JAVA, and this algorithm is helpful to automatic clustering by nonsupervised learning. Finally we prove the value of DASOM in practical application by illustrating applications of customer calling behavior analysis and customer segmentation in telecom industry. Through application in production system of china telecom, we argue basically, compared with existing SOMs algorithms, that clustering algorithm in this paper has feature or advantage of dynamic structure, parameter self adjustment, non-steady data set learning and hierarchical clustering, which is worth further research in study and work in future.