Semantic similarity based feature space optimization using artificial bee colonization

Abstract

Feature selection in text classification has been a popular area of research since decades. It is performed to decrease the learning time of algorithm, increase its classification accuracy and reduce consumption of computational resources. With advancements in the area of Swarm Intelligence and its proven capabilities to solve feature selection problems, this research work is focused on the design and development of a framework for Semantic Similarity based System for Text Classification-Artificial Bee Colony (SSSTC-ABC). The proposed framework provides an efficient mechanism to optimize feature space and performs a classification process taking semantics into account. The proposed framework is tested meticulously using Standard and Non-Standard datasets. This research work uses three classifiers namely Support Vector Machine, Naïve Bayes and K- Nearest Neighbour. The performance evaluation and comparative analysis of evaluation metrics obtained before and after the use of optimization algorithm has been carried out to comprehend the effect of optimization of feature space. This research study was carried out in multiple stages. It began with an extensive review of literature for selection of relevant algorithms, tools and approaches that aided in identification of underlying architecture and design for the proposed framework. This was followed by identification and selection of Standard datasets, their cleaning and structuring for compliance with the framework. The framework was evaluated using a set of metrics after introduction of Swarm Intelligent algorithm ABC. Integration of nature inspired optimization, ABC algorithm in the framework was carried out that resulted in creation of the complete framework. It was tested and evaluated over three Standard and one Non-Standard dataset collections using three classifiers-Support Vector Machine, Naïve Bayes and K-Nearest Neighbour. Statistical analysis of all evaluation metrics was carried out to understand variation in performance with the use of different classifiers.