Thesis (M.Sc.) - Cairo University - Faculty of Graduate Studies for Statistical Research - Department of Information Systems and Technology

Many researches appeared to predict the share price by applying Machine learning algorithm in order to find the most effective prediction model with lowest error percentage. Machine learning has been used in the financial sectors in order to help investors to make better decisions,one of the most important reasons for not reaching a high prediction accuracy is that whoever tries to predict the price in all periods of price change will find that the price changes without laws. In this research the focus is on applying two factors to reduce the randomness of price change, improve the quality of prediction,and reduce the degree of randomness of price movement, and the effect of the two factors will be proven by creating a prediction model before and after applying each factor and comparing the accuracy of the prediction and the percentage of error.Regarding the first factor (abstract the data periods) by Stripping the data from the current price position so that the movement is learned only, regardless of the price place, the goal of applying this work is to make the prediction model learn from the price movement only, regardless of the price position.RegardingThe second factor (filtering the data periods) by Filtering the periods from which the characteristic movement is learned, since we cannot learn from all periods of time series for price, It depends on choosing ideal periods for the price movement, where the price is in a state of saturation for sale or purchase, or record trading volumes, and These argument periods have an expected reaction and a higher degree of expectation

Issued also as CD