#4 TSF: A Time Series Forest for Classification and Feature Extraction

#TSF #classification #timeseries #machinelearning #ml #featureextraction #research #ai

Low computational power is one of the main constraints when building smart devices. To be able to efficiently implement edge AI, one needs to be cognizant of the computational requirements of the different algorithms being explored. Time Series Forests is an efficient ML algorithm for time series classification. It can be potentially be used for many applications involving time series classification, including use cases that involve performing on-device classification using time series data collected from various sensors.

OUTLINE:
0:00 - Intro & Overview
6:42 - Dynamic Time Warping Matching
12:55 - Linear Complexity TSF
19:35 - Related Work (Instance-Based vs Feature-Based)
25:15 - Interval Features
29:08 - Time Series Forest Classifier (Splitting Criteria)
45:51 - Algorithms
51:15 - Time Series Tree and Time Series Forest
55:48 - Temporal Importance Curve
1:06:40 - Experiments
1:18:50 - Computational Complexity
1:21:45 - Conclusion
1:22:18 - QnA and Discussion

Datasets:

Abstract:
A tree-ensemble method, referred to as time series forest (TSF), is proposed for time series classification. TSF employs a combination of entropy gain and a distance measure, referred to as the Entrance (entropy and distance) gain, for evaluating the splits. Experimental studies show that the Entrance gain improves the accuracy of TSF. TSF randomly samples features at each tree node and has computational complexity linear in the length of time series, and can be built using parallel computing techniques. The temporal importance curve is proposed to capture the temporal characteristics useful for classification. Experimental studies show that TSF using simple features such as mean, standard deviation, and the slope is computationally efficient and outperforms strong competitors such as one-nearest-neighbor classifiers with dynamic time warping.

Keywords: decision tree; ensemble; Entrance gain; interpretability; large margin; time series classification;

Authors: Houtao Deng, George Runger, Eugene Tuv, Martyanov Vladimir

Links:
Soham Tiwari

Sahil Khose