Research Article
Concept Drift Detection with Denoising Autoencoder in Incomplete Data
@INPROCEEDINGS{10.1007/978-3-030-94822-1_35, author={Jun Murao and Kei Yonekawa and Mori Kurokawa and Daichi Amagata and Takuya Maekawa and Takahiro Hara}, title={Concept Drift Detection with Denoising Autoencoder in Incomplete Data}, proceedings={Mobile and Ubiquitous Systems: Computing, Networking and Services. 18th EAI International Conference, MobiQuitous 2021, Virtual Event, November 8-11, 2021, Proceedings}, proceedings_a={MOBIQUITOUS}, year={2022}, month={2}, keywords={Concept drift Incomplete data Denoising autoencoder}, doi={10.1007/978-3-030-94822-1_35} }- Jun Murao
Kei Yonekawa
Mori Kurokawa
Daichi Amagata
Takuya Maekawa
Takahiro Hara
Year: 2022
Concept Drift Detection with Denoising Autoencoder in Incomplete Data
MOBIQUITOUS
Springer
DOI: 10.1007/978-3-030-94822-1_35
Abstract
Recent e-commerce and location-based services provide personalized recommendations based on machine-learning models that take into account purchase and visiting histories. Because machine-learning models assume the same distributions between training and test data, they cannot catch up with concept drifts, i.e., changes of behavioral patterns over time. To keep recommendation accurate, it is important to detect concept drifts. Generally, to achieve this, we need complete data (i.e., data without missing values). In real-world datasets, however, there are many incomplete data, and existing concept drift detection techniques do not deal with incomplete data. To address this issue, we investigate how a deep learning technique (denoising autoencoder), which complements missing values, contributes to detecting concept drifts in incomplete data. We conduct experiments on synthetic and real datasets to evaluate the robustness of this technique, and our results show its advantages.
