Automated Assistance for Data Mining Implementation

Abstract

These days there are number of tools available for assistance in implementation of data mining. Some of them are RapidMiner, WEKA, Orange, Rattle, and KNIME. WEKA tool includes a number of techniques like classification, clustering, regression, etc. For solving classification problems this tool provide variety of strategies such as decision tree, neural networks, lazy classifiers etc. For each strategy, the tool allows the user to select specific values for large number parameters for e.g. in case of a neural network classifier, parameters are to be provided by user such as epochs, learning rate, momentum etc. With WEKA an expert user could study which strategy could be the best compatible for the any particular dataset. For that test run can be performed on the test dataset of the user using the various strategies and the resultant outputs are to be evaluated by the expert user its own. This is tedious task to be performed. This paper aims at developing a system that could work on its own even on the evaluation part and thus make it possible for effective implementation of data mining by developing a database to record the nature of data such as number and type of attributes, presence or absence of missing values etc. along with different values for developing classifier models and the accuracy of the classifier. Such a database can then be made available to the novice users to build a model based on past experience.