Predicting Stock Trends: A Multi Agent Approach

This thesis aims to develop a scalable distributed system that provides reliable information about the trend that a certain security will follow in the Stock Market using Machine Learning techniques. The core of the research presented in this thesis is the creation of a Multi Agent based system capable of predicting whether the price of a stock will rise, drop or stay. In order to do so, the problem is divided into three parts: Information Retrieval, Data Analysis and Data Visualization. The first part is focused on retrieving the necessary information from Internet and on transforming this raw data into computer-understandable structures that will allow further analysis. In addition to this, new financial indicators are calculated to provide more meaningful data to the system. The second part is focused on analyzing this preprocessed data using several Machine Learning methods. The methods that have been selected to execute the analysis are: Artificial Neural Networks, Decision Trees, Support Vector Machines and Reinforcement Learning. The idea behind using different methods besides testing the performance of each in this scenario, is the creation of a team of “data analyzers” like the ones found at investment firms. Following the “Keep it Simple” principle, third party libraries have been used when possible to diminish implementation costs. Finally, the third part deals with the problem of how to present the data and results to the users in a clear but informative way. Just this part on this own could perfectly be a Final Project in a Media Degree, so here we will present a gentle introduction to the Data Visualization world. Since this thesis is also a Computer Science Engineering Final Project, emphasis will be made in describing the system architecture and the technologies used to create it. In part because of this, this Thesis aims to create a Proof of Concept for a possible future product instead of realizing just a evaluation of Machine Learning methods applied to predicting stock trend