BrickMIR: A Minimal, Imbalance-tuned, and Ratio-based Framework for Brick Metadata Classification
Year: 2025
Type: conference-paper
Abstract: We propose a multi-label time-series classification framework for building metadata labeling. Rather than relying on absolute values (which vary widely across different buildings), we focus on ratio- and correlation-based features that capture each sensor's relative behavior. Along with minimal data cleaning and labeltuned class weights, this approach generalizes well across buildings with differing baselines, as demonstrated on the Brick by Brick 2024 dataset. Source code is publicly available at https://github.com/chanjunha0/BrickMIR.
Author: Jun Hao Chan
Institution: National University of Singapore
Language: English
Subfield: Signal Processing
Field: Computer Science
Domain: Physical Sciences
Open Access status: gold
Funder: University of New South Wales
OpenAlex ID: W4410637813