Chen Lihui

FEATURE DIMENSION REDUCTION FOR MICROARRAY DATA ANALYSIS USING LOCALLY LINEAR EMBEDDING

Cancer classification is one major application of microarray d ata analysis. Due to the ultra high dimensionality nature of microarray data, data dimension reduction has drawn special attention for such type of data analysis. The currently available data dimension reduction methods are either supervised, where data need to be labeled, or computational complex. In this paper, we proposed to use a revised locally linear embedding(LLE) method, which is purely unsupervised and fast as the feature extraction strategy for microarray data analysis. Three public availab le microarray datasets have been used to test the proposed method. The effectiveness of LLE is evaluated by the classification accuracy of a SVM classifier. Generally, the results are promising.

A novel feature extraction method and hybrid tree classification for handwritten numeral recognition

A hybrid classification system with neural network and decision tree as the classifiers for handwritten numeral recognition is proposed. Firstly a variety of stable and reliable global features are defined and extracted based on the character geometric structures, a novel floating detector is then proposed to detect segments along the left and right profiles of a character image used as local features. The recognition system consists of a hierarchical coarse classification and fine classification. For the coarse classifier: a three-layer feed forward neural network with back propagation learning algorithm is employed to distinguish six subsets {0}, {6}, {8}, {1,7}, {2, 3, 5}, {4, 9} based on the feature similarity of characters extracted. Three character classes namely {0}, {6} and {8} are directly recognized from artificial neural network (ANN). For each of characters in the latter three subsets, a decision tree classifier is built for further fine classification as follows: Firstly, the specific feature-class relationship is heuristically and empirically deduced between the feature primitives and corresponding semantic class. Then, an iterative growing and pruning algorithm is used to form a tree classifier. Experiments demonstrated that the proposed recognition system is robust and flexible and a high recognition rate is reported.

Text document filters using morphological and geometrical features of characters

This paper aims at irregular noise removal in binary document images. A new filtering algorithm called the modified directional morphological filter (MDMF) is introduced with dual properties for eliminating document salt-and-pepper noises, and for remedying eroded character stroke distortion. For eliminating larger noise, an enhanced algorithm called the image geometric structure filter (IGSF) is proposed based on the geometric stroke information of the characters. The IGSF can especially be used for noise removal in alphanumeric documents. Experiments demonstrate that the performance of combining two filters is robust, which can feasibly filter out various isolated noises in the binary images.

Document filters using morphological and geometrical features of characters

Abstract This paper aims at irregular noise removal in binary document images. A new filtering algorithm called modified directional morphological filter is introduced with dual properties for eliminating document salt-and-pepper noises, and for remedying eroded character stroke distortions. For eliminating larger noises other than salt-and-pepper shape, an enhanced algorithm called image geometric structure filter (IGSF) is proposed based on the geometric stroke information of characters. IGSF can especially be used for noise removal in printed/printed–handwritten alphanumeric documents. Experiments demonstrate that the combining performance of two filters is robust, which can feasibly filter out various isolated noises in the binary document images.

An outlier-aware data clustering algorithm in mixture models

A robust mixture model-based clustering algorithm using genetic techniques is proposed in this paper. In many engineering and application domains, noisy samples and outliers often exist in data collections, causing negative effects on performance of data mining methods if they are not made aware of these elements. Classical probabilistic mixture-based clustering is one known to be very sensitive to such situation. To improve its performance, we combine Genetic Algorithm (GA) with the expectation-maximization (EM) procedure of the classical model. When trimmed likelihood is used as fitness function of GA, high representative samples are selected and potential outliers are pruned off effectively during the learning process. Experiments on both synthetic and real data for different applications show that our approach outperforms the classical mixture model, by producing more accurate and reliable results.

Clustering gene data via Associative Clustering Neural Network

We describe a new approach to the analysis of gene expression data using Associative Clustering Neural Network (ACNN). ACNN dynamically evaluates similarity between any two gene samples through the interactions of a group of gene samples. It has feasibility to more robust performance than those similarities evaluated by direct distances. The clustering performance of ACNN has been tested on the Leukemias data set. The experimental results demonstrate that ACNN can achieve superior performance in high dimensional data ( 7129 genes). The performance can be further enhanced when some useful feature selection methodologies are incorporated. The study has shown ACNN can achieve 98.61% accuracy on clustering the Leukemias data set with correlation analysis.

A floating feature detector for handwritten numeral recognition

A novel feature extraction method for handwritten numeral recognition is proposed based on characters geometric structures. A group of stable and reliable global features are defined and extracted. Furthermore, a floating feature detector is proposed to detect and extract tiny segments as fine features. A neural network is employed as the recognisor to conduct experiments on evaluating the feasibility of the new approach. This proposed method demonstrates that the combination of fine features with global features can greatly improve the handwritten character recognition rate compared to those using global features only.

High dimensional gene expression data dimension reduction

Gene expression data analysis is a new approach in cancer diagnosis. Feature selection is an important preprocessing step in gene expression data clustering. In this paper, we demonstrate the effectiveness of feature grouping approach in feature dimension reduction. In our proposed framework, large number of features is grouped to form several feature subsets. By criteria of clustering accuracy, one feature subset is chosen as the candidate subset for further processing by PCA or entropy ranking, and the final feature subset are formed by selecting the features from top ranked ones. Advantage of the framework is that it considers both subset and individual features discrimination power, also it requires little information about the class label. A prototype of the proposed framework has been implemented and tested on the leukemia data set. The results have given positive support to the framework.

K-MEAP: Generating Specified K Clusters with Multiple Exemplars by Efficient Affinity Propagation

Recently, an attractive clustering approach named multi-exemplar affinity propagation (MEAP) has been proposed as an extension to the single exemplar based Affinity Propagation (AP). MEAP is able to automatically identify multiple exemplars for each cluster associated with a super exemplar. However, if the cluster number is a prior knowledge and can be specified by the user, MEAP is unable to make use of such knowledge directly in its learning process. Instead it has to rely on re-running the process as many times as it takes by tuning parameters until it generates the desired number of clusters. The process of MEAP re-running may be very time consuming. In this paper, we propose a new clustering algorithm called KMEAP which is able to generate specified K clusters directly while retaining the advantages of MEAP. Two kinds of new additional messages are introduced in MEAP in order to control the number of clusters in the process of message passing. The detailed problem formulation, the derived updating rules for passing messages, and the in-depth analysis of the proposed K-MEAP are provided. Experimental studies demonstrated that K-MEAP not only generates K clusters directly and efficiently without tuning parameters, but also outperforms related approaches in terms of clustering accuracy.Recently, an attractive clustering approach named multi-exemplar affinity propagation (MEAP) has been proposed as an extension to the single exemplar based Affinity Propagation (AP). MEAP is able to automatically identify multiple exemplars for each cluster associated with a super exemplar. However, if the cluster number is a prior knowledge and can be specified by the user, MEAP is unable to make use of such knowledge directly in its learning process. Instead it has to rely on re-running the process as many times as it takes by tuning parameters until it generates the desired number of clusters. The process of MEAP re-running may be very time consuming. In this paper, we propose a new clustering algorithm called KMEAP which is able to generate specified K clusters directly while retaining the advantages of MEAP. Two kinds of new additional messages are introduced in MEAP in order to control the number of clusters in the process of message passing. The detailed problem formulation, the derived updating rules for passing messages, and the in-depth analysis of the proposed K-MEAP are provided. Experimental studies demonstrated that K-MEAP not only generates K clusters directly and efficiently without tuning parameters, but also outperforms related approaches in terms of clustering accuracy.

Entropy-based robust fuzzy clustering of relational data

Relational data clustering algorithms are proposed to deal with the data represented as the similarity or dissimilarity between each pair of objects. Fuzzy clustering of relational data (FRC) is a recently proposed approach that can handle non-Euclidean distance relational data. Unfortunately, negative values may appear in the clustering process of FRC. Another related algorithm A-P (assignment prototype) applies two different memberships and obtains a more stable minimization procedure. However, the fixed exponent m and sensitivity to initialization make A-P less feasible to some data sets. In this paper, we propose a new entropy-based fuzzy clustering for relational data (EFRC). EFRC and its robust version R-EFRC make use of two types of memberships called partitioning and ranking. Experiments on typical relational data sets and 2-D noisy data sets show that the new algorithm can produce meaningful clustering results and is robust to noise.