International Journal of Computer Science Issues

More than a traditional journal...

Decision trees are the most applicable technique of data mining, because of its power and its simplicity of interpretation. However, learning decision trees from medium to large dataset are different from learning from small dataset, especially when data contain instance that are semantically independent, this lead to lose in accuracy. In our approach, we build patterns according to some criteria with the help of the users knowledge. We use knowledge to filter and reduce the database and remove the data; this is considered as a noise. Learning from the filtered data can generate more accurate and small decision tree. In our experimentation, we show the difference in accuracy between learning over the entire data and filtered data.