1. UKM Learning and Research Repository
2. PHD Thesis / Tesis PHD
3. Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat

Title:	Enhanced feature representation and selection with sentiment for stock market prediction
Authors:	Hani A.K. Ihlayyel (P72276)
Supervisor:	Mohd Zakree Ahmed Nazri, Prof. Dr.
Keywords:	Stock exchanges -- Data processing
Issue Date:	11-Jul-2017
Description:	Stock market prediction using time series has been a common and widely engaged in research activities in the last few decades. As a result of the rapid increase of financial news articles, extracting valuable features become a challenge especially in searching and representing the strong relationship between the news articles and the stock price of the stock market. The financial news article has an enormous amount of features that have a strong impact on stock market prediction. The various data representations lead to discovering the different sequence of pattern to predict the temporal trend of stock prices. This has further encouraged researchers to develop feature selection techniques to understand the relationship between extracted features from the news articles and the stock price. However, the feature representation and selection methods remained simple and do not model all the relative movement and fluctuation of the stock price accurately. The sentiment analysis of news articles also have a significant effect on the stock price movement, but there are issues to model the correlation between the sentiments and overall market that react towards the sudden stock price movements. In order to tackle the above-mentioned drawbacks, this research focus to deal with the limitation of pre-processing steps in feature representation and selection process to discover relationships between the extracted features and stock price. This research proposes three different techniques: (i) A new feature representation algorithm (ELR-BoW) to discover the correlation between the stock prices using different feature representation methods. ii) Feature selection algorithm (AFS) with adaptive characteristic for text categorization that enhances the selection mechanism to in order to capture high-level structures set of time-series features; and iii) Integrating sentiment analysis model (ESA) with stock price movements. To test the performance of the proposed techniques, 8500 articles news article for S&P 500 stock market were used to analyze the stock market. The Bag-of-Words (BoW) technique is implemented to represent the textual information and three feedback measures (PA, DA and CA) were calculated to capture the stock price movements for 20 minutes timeline prediction. In feature representation, The performance of (ELR-BoW) is evaluated by comparing it with the basic BoW technique using Support Vector Machines (SVM) and Naïve Bayes (NB) classifiers, ELR-BoW has improved up to 15.3% in finding the best solution comparing to the basic BoW, and results for the SVM outperform the results for NB in all the comparisons. In feature selection, AFS pushed the f-measure value up to 0.844 and has obtained better results than the basic techniques. The AFS algorithm is further improved by hybridized it with enhanced sentiment analysis (ESA) with ELR-BoW and AFS. The result also shows that the ESA has improved the performance of F-measure value from 0.839 and 0.84 respectively to 0.841 in different feature sizes. This research concludes that there is a significant enhancement to predict movements of stock price to explore the correlation of the extracted features with the stock price. Moreover, the proposed techniques successfully extracts positive patterns of stock market between two sequences when incorporated the structure and unstructured data to explore a strong relation in terms of timeline of 20 minutes, and the terms inside the documents.,Certification of Master's/Doctoral Thesis" is not available
Pages:	184
Call Number:	QA76.9.D343I375 2017 3 tesis
Publisher:	UKM, Bangi
Appears in Collections:	Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.