الفهرس | Only 14 pages are availabe for public view |
Abstract The financial sector is a part of the economy composed of firms and institutions that introduce financial services to commercial and retail customers around the world. This sector includes a wide range of businesses including banks, investment businesses, insurance firms, and real estate organizations. The financial market is a dynamic and composite system where people can transport, buy, and sell currencies, stocks, and derivatives over virtual platforms introduced by agents. Financial stock is a significant factor in the economy. The stock market enables stockholders from owning shares of public companies through the trading process either by exchange or over-the-counter markets. This market has given investors the chance of earning money and having a prospering life through investing small initial amounts of money. The ability to predict stock prices is an important issue regarding the academic area as well as business. Prediction of stock price behavior is an area of strong effect for both academic researchers and industry practitioners, as it is both a difficult task and could lead to increased profits. In this thesis, we developed four approaches for stock analysis and recommendation. The main idea of the first one is to develop a stock price prediction framework using different machine learning models such as and deep learning models. This approach is trained and tested against three stock datasets as Indian, yahoo, and Google stock. In this approach, timeseries stock data problems such as missing values, data duplication, outliers, wide data range, and feature selection are detected and solved by preprocessing techniques, and the overfitting problem is solved using the time series cross-validation technique. The purpose of the second proposed approach is to explore the impact of social news and historical data together on the stock movement and stock trend using an intelligent technique. The main purpose is to develop hybrid modeling that is composed of several stages of random forest classifier and one stage of stacked- LSTM to improve the analysis accuracy. This approach is developed with two versions, the first version using default parameters of intelligent models and the second version adding hyper parameters technique for intelligent models then the proposed approach is improved using time-series cross-validation with 5splitto provide more accurate predictions. By using time series cross-validation, the overfitting problem is eliminated. The third proposed approach was developed to recommend the investor with the best decision about the stock exchange process (good investment/ bad investment) using stock twits’ analysis and historical data analysis during Crises (covid -19 (to predict the future stock price based on intelligent techniques such as random forest and stacked – LSTM and preprocessing techniques in addition to parameters tunning. The main aim of this approach is to explore the impact of StockTwits (customer opinions) and historical data together on the stock close price for future days. Then our system is improved using time series cross-validation to enhance system confidence. from the previous work, some conclusions are gathered. Some work concentrated on using only historical data analysis for the stock price prediction process rejecting the effect of news headlines and stock twits on the future stock movement. Or using the news headlines or stock twits neglecting the effect of historical data analysis. So, in this thesis, proposed hybrid systems are introduced. Also, most of the previous work stopped at the prediction process and does not support the economy system with an efficient recommendation system for the investment decision-making process, this problem is detected and solved using the previous three proposed approaches. Another problem in the previous work there is a lack of stock recommendation systems that are not based on historical stock data analysis, so in the fourth proposed system, we proposed a stock recommendation system for e-commerce that supports the best recommendations without analysis of stock historical data based on stock and investor retail history. The fourth proposed system, a hybrid mixed collaborative and contentbased recommendation system is proposed based on machine learning models and similarity calculation techniques. |