پديد آورنده :
شاكر، محمدحسين
عنوان :
توسعه سيستم معاملاتي در بازار تبادل ارز با استفاده از تكنيكهاي يادگيري عميق و تقويتي
مقطع تحصيلي :
كارشناسي ارشد
گرايش تحصيلي :
مهندسي كامپيوتر- نرم افزار
محل تحصيل :
اصفهان: دانشگاه صنعتي اصفهان، دانشكده برق و كامپيوتر
صفحه شمار :
سيزده، ۷۴ص.: مصور، جدول، نمودار
استاد راهنما :
محمد حسين منشئي
استاد مشاور :
مهران صفاياني
توصيفگر ها :
بازار تبادل ارز , پيش بيني , شبكه هاي عصبي بازگشتي , يادگيري عميق , يادگيري تقويتي
تاريخ ورود اطلاعات :
1397/05/08
رشته تحصيلي :
برق و كامپيوتر
دانشكده :
مهندسي برق و كامپيوتر
چكيده انگليسي :
Developing a Trading System in the Forex Market Based on Deep Neural Networks and Reinforcement Learning Mohammad Hossein Shaker mh shaker@ec iut ac ir May 12 2018 Department of Electrical and Computer Engineering Isfahan University of Technology Isfahan 84156 83111 Iran Degree M Sc Language Farsi Supervisor Dr Mohammad Hossein Manshaei manshaei@cc iut ac ir Abstract The Forex market otherwise known as the foreign exchange market is the center of attention for many people on a dailybasis Due to its high cash flow people think they can make huge amounts of money in short amounts of time But due totheir lack of knowledge and information as well as their emotional behaviors their capital is lost In fact 95 of those whoare active in the Forex market ultimately lose their money In this thesis we are trying to develop two new trading modelsusing the machine learning to predict the future directions of the price movement in the foreign exchange market In the first model using a Recurrent Neural Network the current market conditions and price movements are analyzedto create long or short trade decisions on a single currency pair in a minute by minute fashion The model is capable ofachieving 22 5 in the Kelly risk adjusted return criteria The second model uses a Reinforcement Learning algorithm in combination with Convolutional Neural Networks tochoose the optimal action in the Forex market The main difference between the second model and the first model is that inthe second model in addition to considering the profit or loss of each trade decision the cost of entering into the transactionsis also considered As a result the developed model will be more useful in the real market conditions The model has theability to gain 389 pips of profit over a period of two weeks Also a unique pre processing technique is used to preserve the time series dependencies of price movements in thedeveloped models Key Words Forex Market Prediction Recurrent Neural Networks Deep Learning ReinforcementLearning
استاد راهنما :
محمد حسين منشئي
استاد مشاور :
مهران صفاياني
