Search

Stock investment has become an important investment activity in Taiwan. However, investors usually get loss because of unclear investment objective and blind investment. Therefore, to create a good investment decision support system to assist investors in making good decisions has become an important research problem. Artificial Neural Networks (ANN) can provide relatively good performances in forecasting stock price but it cannot explain the forecasting rules clearly. On the other hand, a decision tree (DT) model can generate some rules to describe the forecasting decisions. This paper focuses on combining ANN and decision trees to create a stock price forecasting model. The experimental result shows that the combined DT+ANN model has 77% accuracy, which is higher than the single ANN and DT models over the electronic industry. [ABSTRACT FROM AUTHOR]

Classification is a data mining (machine learning) technique used to predict group membership for data instances. In this paper, we present the basic classification techniques. Several major kinds of classification method including decision tree induction, Bayesian networks, k-nearest neighbor classifier, case-based reasoning, genetic algorithm and fuzzy logic techniques. The goal of this survey is to provide a comprehensive review of different classification techniques in data mining. [ABSTRACT FROM AUTHOR]

The purpose of this paper is to propose a fuzzy reward function for improving the learning efficiency of reinforcement learning. Reinforcement learning is a sort of online line learning approach. During the learning process, learning system (often called an agent) learns how to operate in the environment, basing upon the effect of action -- reward signal. Actually, the reward signal is always represented with a reward function, of which the role is to evaluate whether the agent acts well or not. According to the distribution of rewards in the space of states, reward function can be classified as two basic forms, dense function and sparse function. Parse function has the advantage of the easy implementation, but its learning efficiency is relatively low. As for dense function, it is very difficult to design when the size of state space is very large. In response to these defects, we present a methodology for designing a fuzzy reward functions by the use of fuzzy inference system. As a result of this solution, not only generalization ability of reward function can be improved, but also we are apt to embed expert experience in reinforcement learning system. Through applying the fuzzy reward function to R-learning and making simulation experiment, we verify that the learning efficiency for R-learning is effectively improved. [ABSTRACT FROM AUTHOR]

Co-training is one of the major semi-supervised learning methods which can induce classifiers from only small number of labeled data and sufficient number of unlabeled data. To perform successfully, co-training requires splitting the features in original data and constructing views. However, it is not so easy to construct good views which satisfy the following conditions for successful performance of co-training: (1) each of views includes good attributes for classification, and (2) the views are sufficiently independent each other. Additionally, some of the recent researches pointed out that co-training can not perform successfully even if the views satisfy the requirements for good performance. We investigate the relationship among the some evaluation criterion for feature splits and the performance of co-training through the well-designed experiments. In this paper, the experimental results and future works for designing ideal artificial feature splits are reported. [ABSTRACT FROM AUTHOR]

This paper analyzes the effects of distance between classes and training datasets size to XCS classifier system on imbalanced datasets. Our purpose is to answer the question whether the loss of performance incurred by the classifier faced with class imbalance problems stems from the class imbalance per se or it can be explained in some other ways. The experiments from 250 artificial imbalanced datasets show that XCS can perform well in some imbalance domains if the training datasets size is large enough and the distance between classes is appropriate. Thus, it dose not seem fair to correlate imbalance datasets directly to the loss performance of XCS. Through this research, we also know what kinds of datasets are suitable for training XCS and dealing with class imbalances alone will not always help improve performance of classifiers. [ABSTRACT FROM AUTHOR]

This paper proposes a new image enlargement method with the estimation of high frequency image components lost in the sampling process. The proposed method is based on a mapping from the low- to high-frequency components, which is constructed by an original image itself that should be enlarged. Concretely, as the mapping system, least squares support vector machine (LS-SVM) which possesses a high generalization ability is employed here. The effectiveness of the proposed method is verified by some image enlargement experiments. Consequently, the experimental results show that the performance of the proposed algorithm is superior to other typical image enlargement algorithms objectively and subjectively. [ABSTRACT FROM AUTHOR]

Clinical (in vivo) 31P MRS is a non invasive protocol for analyzing the energetic metabolism and biomedical changes in cellular level. Evaluation of 31P MRS is important in diagnosis and treatment of many hepatic diseases. In this paper we applied Back-propagation (BP) neural network and support vector machine (SVM) to analyze and visualize 31P MRS for distinguishing hepatocellular carcinoma, hepatic cirrhosis and normal hepatic tissue. Thirty three cases with 66 samples of effective voxel were selected including normal volunteer, cirrhosis and liver cancer patients. Good performance is achieved with limited samples. Experiments prove that SVM model based on 31P MRS offers an alternative and powerful technique for diagnostic prediction of liver cancer in vivo. [ABSTRACT FROM AUTHOR]

Machine learning techniques such as decision tree have been used for network intrusion detection for a decade. These techniques are able to learn the normal and anomalous pattern from training data and generate useful classifiers to detect attacks to computer systems. Most of the current Intrusion Detection Systems (IDS) examine all data features to detect intrusion or misused patterns. Some of the features may be redundant or contribute little (if anything) to the detection process. In this paper we use some KDDCUP99 reduced feature sets, obtained from the previous researches, to classify with decision tree technique. We analyze the performance of using this reduced feature sets for network intrusion detection. We use KDDCUP99 data for training and testing of decision tree. Experiments show that the resulted decision trees have better performance compared to the case in which we use all features. This research discusses the accuracy of decision tree for classifying reduced feature sets. [ABSTRACT FROM AUTHOR]

The XCS classifier system has shown to solve typical classification problems competitively to other machine learning algorithms. This paper is to introduce a flexible classifier selection method in which mach-set generation is performed with respect to the system performance during the training. It also presents a complementary part for the Search Component of the classifier system in order to create new classifiers regarding the input message from the environment, and previous knowledge. The modified XCS learns more reliable and performs obviously better than traditional XCS. [ABSTRACT FROM AUTHOR]

Incorporating background knowledge in the learning process is proven beneficial for numerous applications of logic based learning methods. Yet the effect of background knowledge in graph based learning is not systematically explored. This paper describes and demonstrates the first step in this direction and elaborates on how additional relevant background knowledge could be used to improve the predictive performance of a graph learner. A case study in chemoinformatics is undertaken in this regard in which various types of background knowledge are encoded in graphs that are given as input to a graph learner. It is shown that the type of background knowledge encoded indeed has an effect on the predictive performance, and it is concluded that encoding appropriate background knowledge can be more important than the choice of the graph learning algorithm. [ABSTRACT FROM AUTHOR]