Master of Science / Department of Computing and Information Sciences / William H. Hsu / Stock trading is a complex process which is subject to distinct events both inside and outside an organization. An increase in revenue is a direct influence which causes the stock price to move upwards; likewise, the price of a stock fluctuates due to indirect influences. For example, a stock’s price may move upwards due to a firm arriving at beneficial deals or adding eminent professionals to its board of directors; a rise in correlated stock prices. The stock prices are affected to a great extent by statements of the finance minister and other related officials. In addition, subjective judgments and emotions of traders can also influence the variation of indices and stock prices in the market. The efficient market hypothesis proposes that stock price is unpredictable, assuming all past information has been influenced on current price and therefore it is not useful for the prediction of future price. Nevertheless, there are opposing theories which state that stock prices are predictable through the identification of trends and price patterns based upon past data such as price and volume quotes, balance sheets, and income statements. In the stock market, naive traders (or investors) assume risks due to the above uncertainties, but still have opportunities to make profits through proper, in-depth analysis on sufficient quantities of past data. There are many indicators that are accessible and can help predict the direction of future prices or index values using fundamental and technical data. Fundamental data, derived from the balance sheets and income statements, is preferred for mid-term and long-term investors but not suitable for short-term investors; meanwhile, technical data can be used for short-term investors as well. Technical data is preferred for short-term investments but it can also be used for long-term investments by choosing a specific window of time to look back in determining the indicators for long periods. The current trading model for stocks and indices was developed using an accuracy-based learning classifier system (XCS), which combines reinforcement learning, genetic algorithms, and other heuristics to form an adaptive system whose purpose is to execute stock trades for profit. A test bed developed for experimenting with this system consists of technical data, with candidate features chosen as the most popular indicators.

The growing problem of spam has generated a need for reliable anti-spam filters. There are many filtering techniques along with machine learning and data miming used to reduce the amount of spam. Such algorithms can achieve very high accuracy but with some amount of false positive tradeoff. Generally false positives are prohibitively expensive in the real world. Much work has been done to improve specific algorithms for the task of detecting spam, but less work has been report on leveraging multiple algorithms in email analysis. This study presents an alliance-based approach to classify, discovery and exchange interesting information on spam. Furthermore, the spam filter in this study is build base on the mixture of rough set theory (RST), genetic algorithm (GA) and XCS classifier system. RST has the ability to process imprecise and incomplete data such as spam. GA can speed up the rate of finding the optimal solution (i.e. the rules used to block spam). The reinforcement learning of XCS is a good mechanism to suggest the appropriate classification for the email. The results of spam filtering by alliance-based approach are evaluated by several statistical methods and the performance is great. Two main conclusions can be drawn from this study: (1) the rules exchanged from other mail servers indeed help the filter blocking more spam than before. (2) a combination of algorithms improves both accuracy and reducing false positives for the problem of spam detection.

This thesis has introduced and investigated a new kind of rule-based evolutionary online learning system. It addressed the problem of distributing the knowledge of a Learning Classifier System, that is represented by a population of classifiers. The result is a XCS-derived Learning Classifier System 'XCS with Distributed Rules' (XCS-DR) that introduces independent, interacting agents to distribute the system's acquired knowledge evenly. The agents act collaboratively to solve problem instances at hand. XCS-DR's design and architecture have been explained and its classification performance has been evaluated and scrutinized in detail in this thesis. While not reaching optimal performance, compared to the original XCS, it could be shown that XCS-DR still yields satisfactory classification results. It could be shown that in the simple case of applying only one agent, the introduced system performs as accurately as XCS.