A. Kusiak, Data Mining and Decision Making, in B.V. Dasarathy (Ed.), Proceedings of the SPIE Conference on Data Mining and Knowledge Discovery: Theory, Tools, and Technology IV, Vol. 4730, SPIE, Orlando, FL, April 2002, pp. 155-165. Data Mining and Decision Making Andrew Kusiak Department of Mechanical and Industrial Engineering, The University of Iowa, Iowa City, Iowa 52242 -1527, USA andrew-kusiak@uiowa.edu, http://www.icaen.uiowa.edu/~ankusiak ABSTRACT Models and algorithms for effective decision-making in a data-driven environment are discussed. To enhance the quality of the extracted knowledge and decision-making, the data sets are transformed, the knowledge is extracted with multiple algorithms, the impact of the decisions on the modeled process is simulated, and the parameters optimizing process performance are recommended. The applications discussed in this paper differ from most data mining tasks, where the extracted knowledge is used to assign decision values to new objects that have not been included in the training data. For example, in a typical data mining application the equipment fault is recognized based on the failure symptoms. In this paper, a subset of rules is selected from the extracted knowledge to meet the established decision-making criteria. The parameter values represented by the conditions of this set of rules are called a decision signature. A model and algorithms for the selection of the desired parameters (decision signatures) will be developed. The parameters of this model are updated using a framework provided by the learning classifier systems. Keywords: data mining, temporal data, feature transformation, data transformation, decision making, decision signatures. 1. INTRODUCTION The problems considered in this paper differ from most data mining tasks where knowledge is extracted and used to assign decision values to the new objects that have not been included in the training data. For example, the equipment fault is recognized (i.e., the value of the fault number is assigned) based on the failure symptoms. There are many applications discussed in this paper, where a subset of rules, in particular a single rule, is selected from the extracted knowledge. The parameter values corresponding to the conditions of the rules in this subset are called a decision signature. The decision signature is used to control the process under consideration. One of the questions posed in this research is how to construct the most promising decision signatures for large-scale rule bases. The construction of such decision signatures becomes a challenge due to the temporal nature of the processes from which the data sets considered in this research have been collected. The ideas discussed in this paper are structured in six phases: Phase 1: Data transformation; Phase 2: Rule extraction with alternative algorithms; Phase 3: Decision signature selection; Phase 4: Decision signature validation; Phase 5: A return to the data transformation phase; Phase 6: The acceptance of the decision signature (see Fig. 1). 2. PROBLEM DESCRIPTION The complexity of decision-making in manufacturing, business, and medical applications is rapidly increasing, as the world is becoming data-driven. To cope with this increasing complexity in a changing environment, new modeling and computing paradigms are needed. The salient features of the new modeling and computing paradigm are: G Adaptability of decision-making models to the evolving decision environment.