Stock indicator scanner customization tool

• Main Author: Cheah, Shing Dhee
• Format: Final Year Project / Dissertation / Thesis
• Published: 2022
• Subjects: T Technology (General)
• Online Access: http://eprints.utar.edu.my/4687/1/fyp_CS_2022_CCD.pdf; http://eprints.utar.edu.my/4687/

Nowadays, stock trend prediction has become a famous topic among financial analysts and computer scientists. As developing a model that could accurately predict the directional changes of stock prices will bring huge benefits to investors. The prediction model can act as an assisting tool to helps investors in making trading decisions. Analysing historical prices to make stock prediction is a traditional approach called technical analysis. Technical analysis required investors to identify the trading opportunities of the stock by analysing the historical price manually. To make more accurate prediction in the stock movement, analysis of large amount of historical data is required; however, analysing large amount of historical prices manually by investors is time-consuming and complicated, and this is where machine learning come into handy. Machine learning algorithms able to analyse and extract useful pattern from the huge amounts of data in a short period of time. However, lack of flexibility and dynamicity of the prediction model is the major limit in most previous works. Allowing investors to customize what technical indicators to use in the prediction model is an important factor as experienced investors would know what technical indicators are useful for certain types of stocks. Therefore, this project is proposed to build a stock analysis website that allow investors to customize the inputs of the stock prediction model. This is mainly to combine the power of the machine learning algorithms with the domain knowledge of investors to make more meaningful prediction rather than supply random technical indicators as input of the model like previous work. Since stock prediction problem is considered a time-series problem, long short-term memory (LSTM) will be used as the algorithm in the proposed model, and LSTM has been proven to be good stock prediction model in previous works. LSTM is the improvised version of the recurrent neural network (RNN). Instead of including all past information in the model like RNN, LSTM has the power to include only the useful information by utilizing the gates such as forget gate. This will not only increase the generalizing capability of LSTM model but also solve the vanishing gradient issue in RNN.