Human gait analysis is a well-defined technique for human identification and tracking at distance based on their walking style. It plays an important role in the video surveillance, medical, and defense applications. A number of sensors such as MEMS accelerators, gyroscopes, pressure, and position were deployed for measuring the different gait signals from the body and utilized for the different analysis of human behavior. To effectively reconcile these innovations in medical profession, system is required to identify the most important body features which have an impact on an accurate diagnosis and classification. This study proposes the novel method FOW which intended to choose the best gait features as optimization strategy based on the hybrid integration of whale and firefly algorithms. This approach is utilized for approximating the performance of different classification benchmarks in order to have an efficient medical diagnosis system. In fact, classification issue for the whole set of features is terminated, and it can be significantly pruned. Experimentation has been carried for 35 individuals in which 16 features has been recorded and analyzed. Moreover, the proposed methodology has tested with the different learning algorithms in which integrating with the extreme learning machine has produced nearly 98.5% of accuracy and also outperformed the other existing selection methodologies such as accuracy, sensitivity, and specificity on different classification platforms.

步态分析是一项明确定义的技术，可用于根据步态进行远距离识别和跟踪。它在视频监视，医疗和国防应用中起着重要作用。部署了许多传感器，例如MEMS加速器，陀螺仪，压力和位置，以测量来自人体的不同步态信号，并用于对人类行为的不同分析。为了有效地协调医学界的这些创新，需要系统识别对准确诊断和分类有影响的最重要的身体特征。本研究提出了一种新的方法FOW，旨在基于鲸鱼和萤火虫算法的混合集成，选择最佳步态特征作为优化策略。该方法用于近似不同分类基准的性能，以具有有效的医学诊断系统。实际上，整个功能集的分类问题已终止，可以将其大幅修剪。已经对35个个体进行了实验，其中记录并分析了16个特征。此外，所提出的方法已经用不同的学习算法进行了测试，其中与极限学习机的集成产生了近98.5％的准确度，并且在不同分类平台上也优于其他现有的选择方法，例如准确性，敏感性和特异性。整个功能集的分类问题已终止，可以进行重大修剪。已经对35个个体进行了实验，其中记录并分析了16个特征。此外，所提出的方法已经用不同的学习算法进行了测试，其中与极限学习机的集成产生了近98.5％的准确度，并且在不同分类平台上也优于其他现有的选择方法，例如准确性，敏感性和特异性。整个功能集的分类问题已终止，可以进行重大修剪。已经对35个个体进行了实验，其中记录并分析了16个特征。此外，所提出的方法已经用不同的学习算法进行了测试，其中与极限学习机的集成产生了近98.5％的准确度，并且在不同分类平台上也优于其他现有的选择方法，例如准确性，敏感性和特异性。