Human activity recognition is a problem of classifying activity of a human using accelerometer and gyroscope data recorded by smartphones into well-defined movements. However, it is still a challenging problem to compute a large number of observations in each second and it is also hard to translate the measurements from smartphones into physical activity patterns. In the past few decades, researchers have designed different human activities recognition systems using different algorithms. In this paper, I investigate different approaches to HAR system based on smartphones. I provide a comparison of these approaches and provide recommendations of the best performing solution.