**6. Conclusions**

In this paper, we proposed the eating activity recognition method based on a Bayesian network, using low-power sensors attached to a smartphone and a wrist-wearable device. Contributions of this paper are as follows: (i) obtain and describe the complexity of real activity and limitations of typical learning algorithms using real complex data; (ii) recognize it using only low-power and easily-accessible sensors with low time complexity; (iii) propose the probabilistic model based on the theoretical background; and (iv) provide the various experiments and analysis using large data from 25 different volunteers for 10 activities and various features, showing the usefulness of the proposed method. The proposed method showed an accuracy of 79.71%, which is higher than other learning classifiers, with of 7.54–14.40% better accuracy. We analyzed the error case and the results show that the proposed method could approximately give the answer even when some of contexts or sensor values are very different. Future works include the collection of much larger and representative data, the construction and evaluation of the proposed method for various complex and daily activities, and the evaluation of the proposed method with open data.

**Acknowledgments:** This work was supported by an Electronics and Telecommunications Research Institute (ETRI) grant funded by the Korean government (17ZS1800, Development of self-improving and human-augmenting cognitive computing technology).

**Author Contributions:** Sung-Bae Cho devised the method and guided the whole process to ccreate this paper; Kee-Hoon Kim implemented the method and performed the experiments; and Kee-Hoon Kim and Sung-Bae Cho wrote the paper.

**Conflicts of Interest:** The authors declare no conflict of interest.
