Publication

Abstract : Pattern recognition is the task of choosing the pertinent and descriptive features that best describes the target concept during feature selection (FS). Choosing such descriptive features becomes a daunting task in large-volume datasets which have high dimensionality. In such cases, selecting the discriminative features with better classification accuracy is tedious. To overcome this issue, in recent times, many search heuristics have been used to select the best features from these large-volume datasets. In this work, a sine–cosine algorithm (SCA) with Elitism strategy and new best solution update mechanism is proposed to select best features/attributes to improve the classification accuracy. Improved version of SCA is named as improved sine–cosine algorithm (ISCA). Wrapper-based FS approach is used. ELM with radial basis function kernel is used as the learning algorithm. For experimentation, ISCA is tested with ten benchmark datasets. Experimental results have proved the efficiency of ISCA in achieving better classification performance along with less number of features. Both computational and time complexity has been handled by this algorithm in an expedite manner. The potency of this algorithm is proved by comparing its results with three well-known meta-heuristics such as GA, PSO and basic SCA. Finally, it can be seen that pattern classification using ISCA has been commendable in achieving better classification performance.