Wiphada Wettayaprasit

A novel discretization technique using Class Attribute Interval Average

Discretization algorithm is important for data mining preprocessing because it will help the user to easily understand the data, reduce the complexity of data, reduce processing time, and increase efficiency and accuracy of the data. This paper proposes the new discretization algorithm called Class Attribute Interval Average (CAIA). The algorithm uses 2D-quanta matrix table to calculate each of class individual intervals average and merge the best adjacent intervals to form the new interval. The experimental design uses four-UCI data sets (Iris, Breast Cancer, Heart Diseases, Glass) and four-classification algorithms (J48, RBF, MLP, NB). The comparisons of experimental result with the other six discretization algorithms (EW, EF, ChiMerge, IEM, CAIM, CACC) show that the proposed CAIA has the best mean rank for both of the accuracy and the number of intervals.

Feature Extraction and Interval Filtering Technique for Time-series Forecasting Using Neural Networks

This paper presents the algorithm for feature extraction and interval filtering technique for time-series forecasting using multilayer perceptron neural networks. The algorithm has four parts. The first part is data filtering and interval process. The second part is input feature extraction process from neural networks. The third part is time-series input variables forecasting process. The fourth part is time-series rainfall forecast process. The study uses weather data from the Meteorological Department of Thailand and the United States of America. The experimental results for rainfall forecast receive high accuracy comparing with other methods

Rule extraction from neural networks using fuzzy sets

We present an algorithm for extracting if-then rules from a neural network using fuzzy sets. A set of crisp rules each of which is associated with a certainty factor is initially extracted from a trained neural network. The extraction process is based on fuzzy sets. The crisp rules often induce ambiguity areas in the decision space. The certainty factors of the ambiguity areas are transformed into fuzzy sets. A set of rules with confidence values in natural language terms are then extracted. Experiments using the Iris and the Wisconsin breast cancer databases are used to demonstrate the performance of the method.

Evaluation of fusion techniques for Pan-sharpened THEOS imagery

Pan-sharpened is an image fusion technique, which is designed to increase the resolution of multispectral (MS) images using panchromatic (Pan) image. In this paper, we evaluate fusion techniques for Pan-sharpened THEOS (THailand Earth Observation System) images. A calibrated THEOS imagery with heterogeneous land cover types was evaluated by different fusion techniques. This paper studied both intensity-hue-saturation (IHS) transformation and multiresolution analysis (MRA) to evaluate and improve the sharpening performance. The fusion results were visually and objectively evaluated.

An Automatic Filtering Approach for News Video Feeds on TCP/IP-Based Mobile Devices

Really simple syndication (RSS) technology makes it for users possible to keep track of updated news. Podcasting is the technology based on RSS aims to aggregate multimedia files. For desktop users, viewing redundant RSS news video might not be a problem because of the high efficiency of the machine. But for mobile users, viewing redundant news video in a device with limited resource like PDAs or smart phones is wasting time, memory-space and battery-power. This paper proposes an automatic filtering for news video feeds for use on TCP/IP-based mobile devices. The system aggregates news video feeds from multiple sources and then filters an appropriate video to be shown on users mobile device. To reduce the processing on mobile devices, users are able to access news video through a Web-based reader.

Knowledge Extraction from Self-Organizing Map Using Minimization Entropy Principle Algorithm

Knowledge extraction using self-organizing map produced numeric values. This paper proposes knowledge extraction from self-organizing map using membership function from the minimization entropy principle algorithm to build linguistic intervals. The rough set theory was used in the rule extraction process for the minimum number of rules. The rules were in the form of linguistic if-then rule that user can understand easily. The benchmark data were iris database and Wisconsin breast cancer database. The experimental results received the fewer number of rules with high accuracy

Linguistic Knowledge Extraction from Neural Networks Using Maximum Weight and Frequency Data Representation

This paper presents a method of linguistic rule extraction from neural networks nodes pruning using frequency interval data representation. The method composes of two steps which are 1) neural networks nodes pruning by analysis on the maximum weight and 2) linguistic rule extraction using frequency interval data representation. The study has tested with the benchmark data sets such as heart disease, Wisconsin breast cancer, Pima Indians diabetes, and electrocardiography data set of heart disease patients from hospitals in Thailand. The study found that the linguistic rules received had high accuracy and easy to understand. The number of rules and the number of conjunction of conditions were small and the training time was also decreased

A Noble Feature Selection Method for Human Activity Recognition using Linearly Dependent Concept (LDC)

Human physical activity recognition process using mobile phones is very complicated with many extracted features in which some features are irrelevant or redundant. Removing irrelevant or redundant features is not only reducing the dataset size but also saving the time consuming task. Hence, a reason to pick out the effective and useful features is our main study. We propose a noble feature selection technique using Linearly Dependent Concept (LDC). Our proposed work attempts a new feature selection method on UCI-HAR dataset. For classification, we use the feed forward neural network and compare the performance result with the original dataset. The goal of our study is not only to find an effective and useful features set from the original dataset but also to be better performance than original dataset. Finally, the experimental result of proposed method gives 2.7% more accuracy and reduces the relative error up to 2.67% of the original dataset.

Assistive tool for constructing knowledge structures and suggesting related study materials links

This research proposes a Web-based system for constructing knowledge structures and suggesting study materials links. Within Web-based learning, it is still difficult for learners to identify and choose study materials that match their current and desired abilities. In addition, learners may fail to recognize missing prerequisite learning, and may fail to identify the knowledge they seek. Most e-learning systems do not provide the information needed to assist learners and avoid these difficulties. We propose pedagogically-informed knowledge structures and associated applications, including a tool for designing and building such structures, a tool for navigating the structures for particular purposes (e.g., identifying knowledge missing from learners’ existing knowledge), and a tool for recommending appropriate materials. In this research the knowledge structures are derived from the subject matter within a targeted knowledge domain using task analysis. In this paper, a method of constructing a knowledge structure is proposed. An experimental study investigated users’ design and use of a knowledge structure. Learners will be expected to gain significantly higher levels of achievement by using the knowledge structures and associated tools proposed in this research, and educational communities will be able to share the knowledge structures and use the tools to support learners.

Monitoring, teaching and learning using knowledge maps and structures (MyTeLeMap)

Web-based learning are intended to support learners. It is still difficult, however, for learners to identify and choose study materials that match their current and desired abilities. In addition, learners may fail to recognize missing prerequisite learning, and may fail to understand the the knowledge they seek. Most e-learning systems in Thailand do not provide the information needed to assist learners and avoid these difficulties. This research proposes pedagogically-informed knowledge structures and associated applications, including a tool for designing and building such structures, a tool for navigating the structures for particular purposes (e.g., identifying knowledge missing from learners existing knowledge), and a tool for recommending appropriate materials. Experimental studies will be conducted to validate the design of the knowledge structures and the methods for their construction, and to evaluate the effectiveness of the tools. Learners will be expected to gain significantly higher levels of achievement by using the knowledge structures and associated tools proposed in this research, and educational communities in Thailand will be able to share the knowledge structures and use the tools to support learners.