COOPERATIVE FILTERING: THE ROLE OF HUMAN PERCEPTION ON SELECTING INPUT DATA TO SUPERVISED LEARNING SYSTEMS

摘要

Building a supervised learning neural network to classify domains has always faced the challenge to deal with imperfect set of input data. Independently of the adopted learning method, such as neural networks, agents or statistics, data pre processing is an essential stage for a successful task. Genetic algorithms, variable correlation and exhaustive search are some techniques currently applied to deal with this filtering challenge. The quality of the resulting model depends on the quality of the input data In this paper, we discuss different methods for filtering data emphasizing the power of human intuition. The discussion is developed using a zoology domain of the task to determine liyhofacies in offshore areas to indirectly predict potential oil reservoirs. We present a comparative study using genetic algorithms, correlation coefficients and heuristic intervention (human) applied to lithofacies recognition domain based in well log curves. Our initial results indicate heuristic intervention rarely plays a significant role on data filtering.