Efficient Machine Learning Technique for Web Page Classification

Web page classification plays a major role in information management and retrieval task. Feature selection is an important process for accurate classification of Web pages. Web pages contain several features, and more number of features reduce the classification accuracy. We propose a hybrid feature selection approach which is both efficient and effective for automatic Web page classification problem and also helps the Web search tool to get relevant results in the relevant category. Experiments were conducted by us with various feature selection methods for Web page classification and keyword search problem. From these experiments, it was found that some features present in the initial feature set (IFS) are irrelevant, redundant, and noisy, and they consume more memory space, increase computational time, and give a poor predictive performance. These features can be eliminated using evaluator methods such as principal component analysis, consistency subset evaluator, and search methods such as genetic search and rank search, resulting in minimal and more relevant features. We call these features as intermediate feature set (IMFS), and further optimization in this feature set gives more accurate results. Finally, attribute-selected classifier which is a part of machine learning meta-classifier was applied to the IMFS to get final feature set (FFS), and it was found that accuracy has increased up to 97% and computational time for all classifiers is minimized compared to IFS using WebKb (Faculty and Course) and ODP (Sports) benchmarking datasets. The proposed method yields better classification performance and reduces space requirements and search time in the Web documents compared with the existing methods.