Data-Driven Human Activity Recognition in Smart Environments

Many applications of human activity recognition like healthcare, security etc. show how human activity recognition is important in everyday life. In this paper, we compare different machine learning algorithms like Naïve Bayes (NB), One R (1R) rule, Zero R (0R) rule, J 48 trees, Random Forest (RF) and Random Tree (RT) applied on sensor-based human activity recognition in a home environment. We show that Random Forest achieves better performance in terms of correctly classified instances comparing to other algorithms, while application of 0R rules algorithm achieves significantly the worst performance. Additionally, in order to reduce the dimensionality of the algorithm, we applied wrapper method using the same classifier in the attribute selection. It is shown that using wrapper method the performance of the classification in terms of correctly classified instances is not significantly changed, while in terms of algorithm complexity shows much better performance. After calculating accuracy of each algorithm, we calculate accuracy for each activity classified by each classifier.