IoT Data Repairing with Incremental Multiple Linear Regression

Tao Peng (1) , Sana Sellami (1) , Omar Boucelma (1)

1 DIAMS - Data Integration, Analysis, and Management as Services

Despite the technological improvement on sensors and sensor networks, data emitted by sensors still raise quality issues such as uncertainty or incompleteness, a.k.a missing values. In this paper we address the problem related to IoT data completeness. More specifically, we propose to repair missing values within an IoT data stream, in developing an incremental machine learning method. Our solution processes data as follows: upon arrival of new data at the computer (data) center, our algorithm quickly updates the model after reading again an intermediary matrix instead of accessing historical data. If a missing value is detected, our system will provide an estimation for the missing value based on historical data and the observation of sensors surrounding the one responsible for missing value(s). The paper also presents the performance studies in comparing our approach with existing repairing techniques using real traffic data.

Mots clés: IoT; Data Quality; Linear Regression.

Domaines: Apprentissage [cs.LG]

Tao Peng, Sana Sellami, Omar Boucelma. IoT Data Repairing with Incremental Multiple Linear Regression. BDA 2018 34ème Conférence sur la Gestion de Données – Principes, Technologies et Applications, Oct 2018, Bucarest, Romania. ⟨hal-02092757⟩