Dissertation > Excellent graduate degree dissertation topics show

Research and Implementation of Time Series Classification Based on Semi-supervised Learning

Author: WuLiXia
Tutor: MengJun
School: Dalian University of Technology
Course: Applied Computer Technology
Keywords: Semi-supervised learning Hidden Markov Models Self- training Coordinated training Linear neighbor label passed
CLC: TP181
Type: Master's thesis
Year: 2011
Downloads: 18
Quote: 0
Read: Download Dissertation

Abstract

The time series of widely exist in real life in various fields, including speech recognition, and financial management. The classification of time series is also an important content of the data mining field. The traditional methods of time series sequence main two categories, based on similarity with model-based approach. These classification methods belong to a supervised learning algorithm, requires a lot of mark time series to train in order to get reliable classifier; but it is difficult to get a large number of the tag sequence, and use only the initial sequence of tokens training classifiers obtained classification 's accuracy rate will be very low; while the opposite reality unlabeled time series is very easy to obtain, therefore, the combination of the tag sequence and a large number of unlabeled sequence information to become a hot research training semi-supervised learning classifier method. Sequence classification is based on the time of the semi-supervised learning focus of this paper, for the problem of low classification accuracy of the trained model in the case marked lack of time series based on hidden Markov model (HMM) to study the use of self-training the algorithm iterative learning process to expand tag sequence data sets, and the expanded tagset training HMM training model is more accurate and reliable. Also study the collaborative training algorithm iterative process to expand the set of tags, which coordinated training HMM and the nearest neighbor classifier based classifier will choose, in each iteration, the two base classifier marking data. Mark centralized mistakenly mark, so the research using a method based on rough set upper and lower approximations to edit an expanded set of tags. On the other hand, linear neighbor label transfer semi-supervised learning algorithm to improve the shortcomings constructed between each data point is a close neighbor of Figure rough K-means clustering clustering results data set information on K nearest neighbor neighbors choose to modify and adjust so close neighbor of the diagram of the structure more reasonable. By a large number of comparative experiments on UCR time series data sets, the experimental results show that, higher the HMM classification accuracy expand the set of tags with the Self-Training and Co-Training process training. Synthetic Control, for example, the number of each type of marker for 4:00, the process of using the Self-Training increased 8.11%, and Co-Training 15.19%; improved based on rough K-means clustering (K take 4) LNP is 7.24% higher than the original LNP.

Related Dissertations

CLC: > Industrial Technology > Automation technology,computer technology > Automated basic theory > Artificial intelligence theory > Automated reasoning,machine learning
© 2012 www.DissertationTopic.Net Mobile