Statistically Empirical Integrated Approach for Knowledge Refined Text Classification

Automated text mining is an especially important task in modern data analysis, both from theoretical and experimental points of view. This particular problem has a major interest in the digital age that is related to “Artificial Intelligence, Machine learning and Information Retrieval”. Feature selection and classification of high dimensionality of text data are challenging tasks. In this paper, we adopted an optimal method for dealing with high dimensionality of data. Later, we chose an appropriate strategy (learning algorithm) for an effcient model training. Our empirical evaluation and experimental analysis show that the proposed method performs better compared with other variable selection-based dimension reduction and further text categorisation methods. We exploited several systematic and careful experimentation scenarios in this work to discover what architecture works best for this BBC news dataset. We used 3 hidden layers, each layer with 128 neurons. We observed this architecture optimal as per our specific problem experimentation. Moreover, our proposed method can be useful for improving efficiency and speed-up the calculations on certain datasets.