An Improved Human Activity Recognition by Using Genetic Algorithm to Optimize Feature Vector
Truc D. T. Nguyen, Trung-Tin Huynh, Hoang-Anh Pham · 2018
Nowadays, human activity recognition is emerging as one of the most trending research areas since it supports a wide range of applications, especially in human tracking and sensing. Together with the advancements of embedded devices and sensors, we recognize that smartphones can be exploited as an efficient way to perform human activity recognition since they have become an essential part of humans' daily life. However, the accuracy, execution time and memory consumption are some major challenges in developing a recognition algorithm on smartphones. In this paper, we aim to tackle those challenges by adopting the Genetic Algorithm to optimize the feature vector. The experimental results show that the size of the feature vector is notably diminished using our proposed method. When using the optimized feature vector in machine learning models like Support Vector Machine and 2-Stage Continuous Hidden Markov Models, it has attained high accuracy while the execution time and memory consumption are greatly reduced in comparison with using the non-optimal one.