Knowledge Discovery fromMixed Data by Artificial Neural Network with Unsupervised Learning

Knowledge discovery or data mining from massive data is a hot issue in business and academia in recent years. Real-world data are usually of mixed-type, consisting of categorical and numeric attributes. Mining knowledge from massive, mixed data is challenge. To explore unknown data, visualized analysis allows users to gain some initial understanding regarding the data and to prepare for further analysis. Self-organizing map (SOM) has been commonly used as a visualized analysis tool due to its capability of reflecting topological order of the high-dimensional data in a low-dimensional space. Interesting patterns can thus be discovered by visual clues, possibly leading to discovery of valuable knowledge. In previous studies, an extended SOM has been proposed to visualize mixed-type data. However, the model works under the setting of supervised learning in order to measure the similarity between categorical values. In this article, we propose a model which can work under the setting of unsupervised learning so that neither class attribute nor domain expert is required. Experimental results are reported to demonstrate effectiveness of the proposed approach.