A wrapper metaheuristic framework for handwritten signature verification

Handwritten signature verification is one of the most fast, intuitive and cost-effective tools for user authentication. However, the analysis of the handwritten signature may become complicated due to the large dimensionality. Feature selection is treated as one of the most commonly applied preprocessing techniques in data mining to deal with the dimensionality. Since the feature selection can be considered as an optimization problem, we propose a wrapper metaheuristic framework for the handwritten signature verification, which uses the k-nearest neighbor classification algorithm to guide a metaheuristic optimization algorithm in order to find the best subset of features. The following metaheuristics are used in the work: swallow swarm optimization (SSO), binary dragonfly swarm optimization (BBA), population algorithm with memory (PAO), genetic algorithm (GA), binary bat algorithm (BBA), binary whale optimization algorithm (WOA) and particle swarm optimization (PSO). The experiments are carried on the SVC2004 dataset by comparing the employed metaheuristics with each other from different perspectives, including the classification performance, the relatives frequencies of selected features and the feature subset size. According to results, the proposed wrapper framework obtains extremely promising results for handwritten signature verification. In particular, the proposed framework can increase the classification accuracy at least or more than 8%, even though selecting between 2 and 6 features among 100 features in most cases. It is also observed that the proposed framework with PSO can achieve slightly better results than with the other metaheuristics in most cases. It can therefore be suggested that the traditional metaheuristics like PSO and GA maintain their position and entrench their popularity in a wide range of fields despite a variety of the recently introduced metaheuristics.