We introduce a new focus for data mining, which is concerned with knowledge discovery in image databases. We expect all aspects of data mining to be relevant to image mining but in this first work we concentrate on the problem of finding associations. To that end, we present a data mining algorithm to find association rules in 2-dimensional color images. The algorithm has four major steps: feature extraction, object identification, auxiliary image creation and object mining. Our algorithm is general in that it does not rely on any type of domain knowledge. A synthetic image set containing geometric shapes was generated to test our initial algorithm implementation. Our experimental results show that image mining is feasible. We also suggest several directions for future work in this area. 1

One of the main obstacles in applying data mining techniques to large, real-world databases is the lack of efficient data management. In this paper, we outline a two-level architecture, consisting of a mining tool and a database server. Key elements in its success are a clear separation of concerns: the mining tool organizes and controls the search process, while all data-handling is performed by the parallel main memory DBMS. Data is stored as a set of binary tables. The interaction consists of queries for statistical information. Properties of the DBMS and the search algorithm are exploited for optimization of the data handling. In particular, results of previous computations are re-used, and I/O activity is reduced by keeping a small hot-set of binary tables in main-memory. As test results show, this system handles large datasets at a competitive performance. CR Subject Classification (1991): Data storage representations (E.2), Database systems (H.2.4) parallel systems, quer...

We describe some initial experiences in dealing with the task of acquiring knowledge where a very large collection of case histories is available. A Knowledge Discovery in Databases (KDD) approach is taken. KDD is the process of extracting novel information and knowledge from large databases, consisting of many interacting stages performing specific data manipulation and transformation operations with an information flow from one stage onto the next (and usually with feedback into previous stages). We characterise our experiences of this process for the task of acquiring knowledge for the domain of motor vehicle insurance premium setting for NRMA Insurance Limited. Keywords: Knowledge acquisition, knowledge discovery in databases, data mining, insurance premiums, risk analysis, fraud. 1 Introduction Knowledge Discovery in Databases (KDD) is commonly defined as the "non-trivial process of identifying valid, novel, potentially useful, and ultimately understandable patterns in data" (...