Financial numeric and textual data based stock prediction using machine learning techniques /
Source of web-text and numerical data analysis for stock prediction is the challenging tasks in today's stock market engineering. Day traders in the stock market face the common issues of decision making, which is mostly dependent on daily or weekly bases data analysis. To overcome this problem...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
Kuala Lumpur :
Kulliyyah of Information and Communication Technology, International Islamic University Malaysia,
2020
|
| Subjects: | |
| Online Access: | http://studentrepo.iium.edu.my/handle/123456789/10425 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Source of web-text and numerical data analysis for stock prediction is the challenging tasks in today's stock market engineering. Day traders in the stock market face the common issues of decision making, which is mostly dependent on daily or weekly bases data analysis. To overcome this problem, web text mining and data mining analysis techniques applied on stock market closing values, which brings the most technical approach. In terms of stock-market textual data classification, the applicable soft-computing technique holds two classifications as binary and multinomial with clustering algorithms used to apply for analysis. New prediction models overcome the drawback of previous research and indicate the necessity of classification by creating prediction algorithm with a token or a polar based financial text weighting scheme of intensive scale value (ISV) system. Binary classification helps to improve the sense of positivity and negativity with intensive value to evaluate the vast amount of financial textual data for trading decision. This research improves with the technical correlation that addressed the problem of categorical financial textual and numerical data throughout various soft-computing techniques. Targeted numerical and textual data rely on subsequently neural network, binary and multinomial classification to improve the prediction techniques by feature engineering. In terms of textual data, the novel financial data dictionary is prepared based on Harvard reference weighting schemed valued that defined as a likely result in new Stock Prediction Model. Financial news-based text analysis techniques improve the classification scenario with Naïve Bayes binary classification through financial data dictionary. Beside the text analysis, the feed-forward neural network architectural model also improved based on backpropagation neural network structural that approached by defining the correlation between the actual and prediction trend of a daily basis. Daily stock price prediction is the main objective of this research and very essential to generate accurate prediction through online daily basis financial news data. The new architectural neural network model performs with sequential data as hidden with the dataset which applied by the multi-objective optimization algorithm. Throughout feature engineering, setting by scaling value determines the weight factors of developing a neural network that used to define more precious trend within this model. This model enabled to calculate the highest frequent value that occurred on a large dataset, and clustering indicate the stock trend of the prediction. Based on the numerical financial data, new Stock Prediction Model (SPM) have developed for analyzing market movement from two benchmarks numerical stock market dataset those are S&P 500index and OHLCV dataset. Developing integrated classification techniques conducting with prediction analysis based on its classification accuracy as defined in this research 82% which is obvious and better than previous research. The performance with feature engineering in text classification also gain 93%, whereas multilevel and binary classification have performed as combined to gain the best accuracy level. Performance of the proposed approach was estimated by evaluating various parameter as part of the information retrieval field, as seen in this experimental result. However, developing model impacts on academical research philosophy in terms of financial data classification but not highly recommend using in real trading analysis. |
|---|---|
| Item Description: | Abstracts in English and Arabic. "A thesis submitted in fulfilment of the requirement for the degree of Doctor of Philosophy in Computer Science." --On title page. |
| Physical Description: | xviii, 226 leaves : colour illustrations ; 30cm. |
| Bibliography: | Includes bibliographical references (leaves 194-209). |