INTERNATIONAL JOURNAL OF SCIENTIFIC & TECHNOLOGY RESEARCH VOLUME 9, ISSUE 02, FEBRUARY 2020 ISSN 2277-8616 4074 IJSTR©2020 www.ijstr.org Arabic Financial Stock Prediction Approach Based On Deep Learning Technique Maryam Mohamad Al-Zawahra, Mohammad Bany Baker, Mohammad Zayed Almuiet Abstract: The prediction of the financial stock market based on the published financial news is considered to be a major issue in the financial field, particularly in light of the significant contents of the news article. Nevertheless, the prediction technique used to gather useful terms from the Arabic financial news has been elusive and is yet to be established. Therefore, this study aimed to develop and propose a practical solution by which the Arabic stock market can be predicted through the analysis of event terms from the Arabic financial news articles. The approach obtained 89.68% accuracy in such prediction, indicating the usefulness of the approach. This is a unique and pioneering study that applies machine learning method to predict Arabic stock market. The study has several implications to the financial field and profession. Index Terms: Financial stock market, Prediction, Machine learning, Deep Learning, Recurrent Neural Network. ——————————  —————————— 1 INTRODUCTION Stock market is a crucial component of the financial market, in terms of the interactions and dealings that take place through it. Investor and speculator circles in the market are looking to raise profitability by the analysis of the information provided concerning the market. In this regard, one of the main market information sources is the financial news articles and such articles have been extensively utilized for analysis among investors [1]. In the current era of data, referred commonly as the Big Data Era, news articles have shown a monumental increase, and faced with a considerable pile of news articles, institutions have been increasingly turning to high processing computer power for analysis. Support systems predictions can help investors sift through noises and reach wise and informed decisions as contended by [2]. Therefore, this study proposes a model to conduct news articles analysis for accurate predictions to solve the issue. 1 Moreover, predictive measures can basically be divided into two major analyses, which are technical and fundamental analysis. They are distinct from each other according to the data input, and also, the former is used with historic market data, while the latter is used for news concerning country, society, firm, and the like. Majority of prior literature has conducted technical analyses based on quantitative historic market data and the traders‘ inclination towards the use of technical qualitative techniques. In relation to this, basic data is more difficult to use as the data input, particularly if it lacks structure. However, basic data may be obtained from structured and numerical sources such as macro-economic data or regular financial reports provided by the governments and their banks. Such basic data aspect has not been extensively studied, although authors do rarely illustrate their predictive value (e.g., [3]). Basic data in the form of unstructured text is the most difficult aspect to examine and is thus, the direction that the study adopted. This may be exemplified by the textual information available in social media, blogs, forums and news [3]. In the field of technical analysis, researchers have primarily made use of mathematics for the analysis of historical stock price patterns, and eventually predict stock prices for the future. Many algorithms have been adopted for such purpose in different forms; for instance, multiple kernel learning in [4], deep learning in [5] and [6], stepwise regression analysis in [7], among others. Despite the clear and consistent outcomes achieved, it remains challenging to predict stock prices in an accurate manner through the sole use of historical prices owing to the unpredictable events that can influence such prices. Meanwhile, in [8] study, the authors employed fundamental analysis of natural language processing (NLP) in their attempt to analyze financial news and statements from the firm, with the ultimate aim of predicting future stock trends (uptrend and downtrend). Specifically, NLP‘s bag-of-words technique is the most often used method for the features extraction from the news articles, measuring each word‘s presence and transforming text information into vector spaces with their help. Following this step, machine learning algorithms are used to establish the relationship between word patterns and the movements of stock prices. Regardless of the accurate ability of bag-of-words technique in several studies (e.g., [5], [8]), one crucial element has been left out in the prediction of directions and that is the sentiment of the article. In relation to this, a specific phase in concerns the interpretation of the investors of the published news articles after which they are transformed into positive/negative sentiments to reach decisions as to whether stocks should be purchased, held or sold on the basis of the interpretations of sentiments. Lastly, the aggregate of the market prices is used on each investor‘s actions, after which, this is represented in the final price trends. In other words, based on the above, the combination of sentiment analysis and natural language processing could prove effective. Lastly, machine learning algorithms are used to obtain information on the relationship between extracted news features and stocks trends. More recently, Deep Neural Networks (DNNs) – a category of machine learning – has garnered success after another in different fields, for instance, in speech recognition in [9] and computer vision in [10]. In this line of study, neural networks have become popular owing to the image-based applications ___________________________________ • Maryam Mohamad Al-Zawahra is currently an lecturer in Department of Software Engineering, Faculty of Prince Al-Hussein bin Abdullah II of Information, Technology, Hashemite University, Zarqa, Jordan. E- mail: maryam_alz@hu.edu.jo • Mohammad Bany Baker is an Assistant professor at Center for Artificial Intelligence Technology, Faculty of Information Science and Technology, Universiti Kebangsaan Malaysia, Bangi, Malaysia. E- mail: mohammad_banibakr@yahoo.com • Mohammad Zayed Almuiet is an Assistant professor at Center for Artificial Intelligence Technology, Faculty of Information Science and Technology, Universiti Kebangsaan Malaysia, Bangi, Malaysia. E- mail: malmuiet@gmail.com