Automatic selection and evaluation on data mining algorithms

For traditional data mining tasks, algorithms are commonly selected by manual effort. However, it is a challenge for any practitioner to select the most appropriate algorithm from hundreds of candidates. To address this issue, we have proposed a novel model for supporting automatic selection on data mining algorithms. The model incorporates the extracted characteristics of data sets and the dynamically established rule sets into the procedures of automatic algorithm selection, which significantly accelerates the progress of algorithm se lection for a variety of data mining tasks. In addition, we have investigated a set of quantized and subdivided evaluation criteria for supporting high quality algorithm selection. Experimental work has been conducted to ve rify the feasibility and effectiveness of the proposed model.