Cheng Hao Jin

A New Ensemble Method with Feature Space Partitioning for High-Dimensional Data Classification

Ensemble data mining methods, also known as classifier combination, are often used to improve the performance of classification. Various classifier combination methods such as bagging, boosting, and random forest have been devised and have received considerable attention in the past. However, data dimensionality increases rapidly day by day. Such a trend poses various challenges as these methods are not suitable to directly apply to high-dimensional datasets. In this paper, we propose an ensemble method for classification of high-dimensional data, with each classifier constructed from a different set of features determined by partitioning of redundant features. In our method, the redundancy of features is considered to divide the original feature space. Then, each generated feature subset is trained by a support vector machine, and the results of each classifier are combined by majority voting. The efficiency and effectiveness of our method are demonstrated through comparisons with other ensemble techniques, and the results show that our method outperforms other methods.

Correlation of Amino Acid Physicochemical Properties with Protein Secondary Structure Conformation

Eight representative physicochemical properties of amino acids are considered to encode each residue and correlative information is examined in relation to the formation of protein secondary structure. Features salient at the coarse level are first gleaned through vector quantization technique and then more refined class-specific features are identified based on the vector element-wise analysis. Effectiveness of the method has been validated in experiments to predict secondary structure using the widely used protein sequence sets. Heuristic rationale for advantage of using physicochemical properties of amino acids over the conventional statistics-based method relying on the frequency of residue occurrence is also presented.

Ensemble method for classification of high-dimensional data

Ensemble methods, also known as classifier combination were often used to improve the performance of classification. Growing problem of data dimensionality makes a various challenges for supervised learning. Generally used classification methods such as decision tree, neural network and support vector machines were difficult to be directly applied on high-dimensional datasets. In this paper, we proposed an ensemble method for classification of high-dimensional data, with each classifier constructed from a different set of features determined by partition of redundant features. In our method, the redundancy of features was considered to divide the original feature space. Then, each generated feature subset was trained by support vector machine and the results of each classifier were combined by the majority voting method. The efficiency and effectiveness of our method were demonstrated through comparisons with other ensemble techniques, and the results showed that our method outperformed other methods.

Application of calendar-based temporal classification to forecast customer load patterns from load demand data

We present temporal classification technique in this paper how to predict power load patterns from load demand data. Since the power load patterns have time-varying characteristic and very different patterns according to the hour, time, day and week and so on, it gives rise to the uninformative results if only traditional data mining is used. Therefore, we propose a temporal classification method for forecasting electrical customer load patterns. The main tasks include cluster analysis and temporal classification technique. Cluster analysis is used to create load pattern classes and the representative load profiles for each class. Next, the classification method uses representative load profiles to build a classifier able to assign different load patterns to the existing classes. The proposed classification method uses the calendar-based temporal expression to discover load patterns in multiple time granularities such as short-, mid-, and long-term time interval. Lastly, in order to show the feasibility of temporal classification technique, the proposed methodology is applied on a set of high voltage customers of the Korea power system, and the results of our experiments are presented.

Classification of Closed Frequent Patterns Improved by Feature Space Transformation

In some real-world applications, the predefined features are not discriminative enough to represent well the distinctiveness of different classes. Therefore, building a more well-defined feature space becomes an urgent task. The main goal of feature space transformation is to map a set of features defined in a space into a new more powerful feature space so that the classification based on the transformed data can achieve performance gain compared to the performance in the original space. In this paper, we introduce a feature transformation method in which the feature transformation is conducted using the closed frequent patterns. Experiments on real-world datasets show that the transformed features obtained from combining the closed frequent patterns and the original features are superior in terms of classification accuracy than the approach based solely on closed frequent patterns.

Design of Context Analysis System on USN Environment

With advances in sensor techniques, much research is focused on USN (Ubiquitous Sensor Network) computing services and context analysis service is an important field among them. However, sensor stream data, which is generated from sensors used in many USN domain applications, is too fast to control each of them and the volume is too huge to store the whole data. Hence, in order to provide rapid and reliable context analysis service over such sensor stream data, a context analysis system which supports the functionality of sliding window is needed. The context model used in this system is a WHEN-DO context analysis model. This context analysis model is designed to be used as follows: If the sensor stream data satisfies condition in ‘WHEN’ clause, then it will execute actions specified in ‘DO’ clause in WHEN-DO context analysis model. Hence, this system also works as the same way. First determine if the received sensor stream data satisfies condition in ‘WHEN’ clause, if the condition is satisfied with the value of received sensor stream data, then it will execute action in ‘DO’ clause in WHEN-DO context analysis model. Using WHEN-DO context analysis model in this system can take corresponding actions according to user defined conditions. Our proposed context analysis system can be applied to many other USN environment applications such as monitoring the status of a building and then taking actions to that condition.

Detection of Correlated Microarray Expressions Using Difference Values

We present a multivariate method to find genes with correlated expressions across the samples. Our contributions in this study are three-fold: firstly, we develop a difference vector-based technique which unfolds hidden correlations over a subset of genes, secondly, we present a similarity measure which enables grouping of gene expressions based on local similarity regardless of global distance, and thirdly, we devise visualization tools that are useful for conducting an ‘explainable’ analysis. Integrating these techniques with the spectral clustering algorithm, biomarker genes can be effectively identified. We have evaluated our method on six microarray datasets that are widely used as a testbed. When we apply our method in the sample classification problem as well as gene selection, we can successfully explain the source of misclassification by showing the correlation patterns for a subset of genes with the aid of the visualization tools.

A New Approach for Calculating Similarity of Categorical Data

Similarity measure is very important in data mining techniques such as clustering, nearest-neighbor classification, outlier detection and so on [1][4]. There are many similarity measures have been proposed. For numeric data, there are many Minkowski distance-based similarity measures. However, the similarity measures for categorical data have been studied for a long time, it also has many issues. The main issue is to understand relationship between categorical attribute values. For categorical data, the similarity measure is not clear as well as numeric data. In this paper, we propose a new approach to understand relationship between categorical data. This approach is based on artificial neural network to extract significant features for computing distance between two categorical data objects.