© 2020, Springer-Verlag London Ltd., part of Springer Nature. Genetic algorithm (GA) is a nature-inspired algorithm to produce best possible solution by selecting the fittest individual from a pool of possible solutions. Like most of the optimization techniques, the GA can also stuck in the local optima, producing a suboptimal solution. This work presents a novel metaheuristic optimizer named as the binary chaotic genetic algorithm (BCGA) to improve the GA performance. The chaotic maps are applied to the initial population, and the reproduction operations follow. To demonstrate its utility, the proposed BCGA is applied to a feature selection task from an affective database, namely AMIGOS (A Dataset for Affect, Personality and Mood Research on Individuals and Groups) and two healthcare datasets having large feature space. Performance of the BCGA is compared with the traditional GA and two state-of-the-art feature selection methods. The comparison is made based on classification accuracy and the number of selected features. Experimental results suggest promising capability of BCGA to find the optimal subset of features that achieves better fitness values. The obtained results also suggest that the chaotic maps, especially sinusoidal chaotic map, perform better as compared to other maps in enhancing the performance of raw GA. The proposed approach obtains, on average, a fitness value twice as better than the one achieved through the raw GA in the identification of the seven classes of emotions.