Hybrid Bat and Salp Swarm Algorithm for Feature Selection and Classification of Crisis-Related Tweets in Social Networks

Twitter is a useful tool for effectively tracking and managing crisis-related incidents. However, due to many irrelevant features in textual data, the problem of high dimensionality arises, which eventually increases the computational cost and decreases classification performance. Thus, to handle such a problem, this work presents a Spark-based hybrid binary Bat (BBA) and binary Salp swarm algorithm (BSSA) named SBBASSA for feature selection and classification of crisis-related tweets. In the proposed technique, the hybridization of standard BBA and BSSA algorithms is performed to enhance their exploration capabilities, then the combined algorithm is implemented in parallel using Apache Spark framework to reduce the overall execution time during the feature selection process. A support vector machine (SVM) classifier is applied during the wrapper-based feature subset selection and classification. The performance of the proposed SBBASSA was analyzed on six benchmark crisis tweet datasets, namely Hurricane Sandy, Boston Bombings, Oklahoma Tornado, West Texas Explosion, Alberta Floods, and Queensland Floods, and then compared with standard BSSA, BBA, and binary particle swarm optimization (BPSO). Results showed that SBBASSA performed competently in the feature selection and classification, outperformed other algorithms in crisis tweet classification, and achieved the highest accuracy with the lowest feature set in a reduced execution time.