Interactive data mining from clinical inspection data

In this paper, we use Interactive Evolutionary Computation in order to select relevant features in inductive learning for Data Mining tasks. The method we have proposed is used to discover efficient decision knowledge from noisy inspection data in a medical domain. This paper describes the principles SIBILE (Simulated Breeding and Inductive Learning), which we have developed for practical data mining problems, then we applied SIBILE to a data set on clinical patients about liver functions. Our approach to the problem is characterized as follows: (1) Repeat an apply-and-evaluate loop of C4.5 inductive learning programs by a human expert with medical knowledge to assess the performance of the program; and then (2) Apply our GA-based feature selection method with human-in-a-loop interactive manner. An described elsewhere, SIBILE has shown good performance in marketing decision making problems. The contribution of this paper is to demonstrate that the proposed method is so powerful that SIBILE is applicable to more complex and severe problems In medical task domains.