Time Series Feature Extraction and Selection Tool for Fire Data
Jun Wang, Youwei Jia, Eugene Yujun Fu, Jiajia Li, Wai Cheong Tam
This paper aims to facilitate the use of machine learning to carry out supervised classification/regression tasks for time series data in fire research. Specifically, a feature engineering tool, FAST (Feature extrAction and Selection for Time-series), is developed. Using hypothesis test method together with principal component analysis, relevant features with high significance to the prediction are selected. A study case is presented for the use of FAST. Results demonstrate the importance of obtaining effective features and its potential benefits. It is expected that the feature engineering tool can help scientists and engineers in the fire research community develop accurate machine learning based models.
17th International Conference on Automatic Fire Detection (AUBE 20) and Suppression, Detection
and Signaling Research and Applications Conference (SUPDET 20)
September 15-17, 2020
Mülheim an der Ruhr, -1
Time series, feature extraction, feature selection, machine learning