Year: 2025
Type: conference-paper
Abstract: We present a machine-learning-based approach to efficiently classify IoT device data in the AIcrowd | Brick by Brick 2024 competition, enabling automation in smart building management. The increasing prevalence of IoT devices in modern buildings has led to an increase in time-series data from diverse sources. Efficiently classifying and organizing this data according to an ontology is crucial for applications such as energy optimization, fault detection, and predictive maintenance. However, manual classification is costly and time-consuming. To address this, we propose a Quartile-Based Shape Segmentation (QBaSS) method, which extracts lower-dimensional time-series representations for stratified classification using gradient boosting. Our results demonstrate reliable classification for most device types, achieving a Macro F1 score of 0.509 with the test data.
Institution: Flame University
Language: English
Subfield: Signal Processing
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W4410636499