Improved Social Network Search based Twin Extreme Learning Machine

Abstract An excellent classifier termed as twin extreme learning machine (TELM) has been widely used in semantic classification, medical image recognition and other fields. However, due to the random weights and biases are generated in the feature space of TELM, there will be some non-optimal or unnecessary weights and biases, which will reduce the efficiency of the TELM and increase the testing cost. In this paper, we first propose a novel metaheuristic algorithm called improved social network search (ISNS) to solve optimization problems. ISNS introduces logistic chaotic mapping and adds a reverse learning strategy of lens principle on the Social Network Search (SNS), which can accelerate the convergence of SNS. Then, in order to optimize the weights and biases of TELM we propose an improved social network search based twin extreme learning machine algorithm (ISNS-TELM), which can improve TELM performance effectively. Finally a number of mathematical functions were chosen to evaluate the performance of our ISNS algorithm. Furthermore, we compare our ISNS-TELM with some state-of-the-art algorithms on the UCI dataset and several multi-category imagine dataset. According to the experiment results, the ISNS-TELM is capable of achieving better training results and getting faster convergence speed.