Intelligent Database Systems Lab Presenter: HONG, CHIA-TSE Authors: Yen-Hsien Lee, Chih-Ping Wei, Tsang-Hsiang Cheng, Ching-Ting Yang 2012. DSS Nearest-neighbor-based.

Intelligent Database Systems Lab

Presenter: HONG, CHIA-TSE

Authors: Yen-Hsien Lee, Chih-Ping Wei, Tsang-Hsiang Cheng, Ching-Ting Yang

2012. DSS

Nearest-neighbor-based approach to time-series classification

Intelligent Database Systems Lab

Outlines

Motivation Objectives Methodology Experiments Conclusions Comments

2

Intelligent Database Systems Lab

Motivation• Prior classification analysis research predominately

focuses on constructing a classification model from

training instances that involve nontime-series

attributes.

• Traditional classification analysis techniques such

statistical-transformation-based approach often results

in information loss and, in turn, imperils classification

effectiveness.

(55, 45, 35, 25, 15) ( 5, 20, 35, 50, 65)

Intelligent Database Systems Lab

Objectives• This study aims to propose and develop a novel time-

series classification technique based on the k-nearest-

neighbor (kNN) classification approach.

• The preservation of trends in time-series sequences

when inducing a classification model for a time-series

classification problem can reduce information loss.

Intelligent Database Systems Lab

Methodology(review): Analysis and selection of learning strategy for time-series classification

5

• Model-based learning strategy

• Instance-based learning strategy

Intelligent Database Systems Lab

Methodology - kNN-based time-series classification technique

Decision combination methods

6

Time-series similarity measureKNN-TSC

Intelligent Database Systems Lab7

Experiments

Intelligent Database Systems Lab8

Experiments• Performance benchmark

Intelligent Database Systems Lab9

Experiments• Parameter tuning experiments

Intelligent Database Systems Lab10

Experiments• Comparative evaluation

Intelligent Database Systems Lab

Conclusions

• The empirical results show that the proposed kNN-TSC

technique achieves better performance than the traditional

statistical-transformation-based approach does.

• With the use of the stratified average method for decision

combination, kNN-TSC technique can effectively handle the

asymmetric class-distribution problem.

Intelligent Database Systems Lab

Comments

• Advantages- Achieves better performance.

• Applications- Time-series classification problems.