Machine Learning based Outlier Detection in IoT Greenhouse

Data Monitoring becomes mandatory for several IoT applications including smart greenhouse. It aims to increase the quality of information by identifying existing errors and anomalies, especially using outlier detection process by machine learning-data classification. Existing approaches require the knowledge of data characteristics in advance. However, this requirement is not always possible in IoT due to the heterogeneity of devices. Therefore, this paper provides VoteIoT: a new monitoring method based on data analytic and vote clustering outcome. In this way, we increase the probability of making a good decision and guaranteeing a good harvest of greenhouses. To evaluate the proposed solution, we used a real database extended by augmented data. The results show a good response time, below 0.01 seconds, as well as a good detection accuracy of 97%. Additionally, the false alarms are below 3% and therefore, a low useful data loss. Moreover, we have managed to increase the probability of a good decision compared to existing solutions.