Wyniki wyszukiwania - BazTech

1

Using Diversity for Classifier Ensemble Pruning : an Empirical Investigation

Ahmed M. A. O., Didaci L., Lavi B., Fumera G.

Theoretical and Applied Informatics

|

2017

|

Vol. 29, No. 1-2

25--39

EN

The concept of `diversity' has been one of the main open issues in the field of multiple classifier systems. In this paper we address a facet of diversity related to its effectiveness for ensemble construction, namely, explicitly using diversity measures for ensemble construction techniques based on the kind of overproduce and choose strategy known as ensemble pruning. Such a strategy consists of selecting the (hopefully) more accurate subset of classifiers out of an original, larger ensemble. Whereas several existing pruning methods use some combination of individual classifiers' accuracy and diversity, it is still unclear whether such an evaluation function is better than the bare estimate of ensemble accuracy. We empirically investigate this issue by comparing two evaluation functions in the context of ensemble pruning: the estimate of ensemble accuracy, and its linear combination with several well-known diversity measures. This can also be viewed as using diversity as a regularizer, as suggested by some authors. To this aim we use a pruning method based on forward selection, since it allows a direct comparison between different evaluation functions. Experiments on thirty-seven benchmark data sets, four diversity measures and three base classifiers provide evidence that using diversity measures for ensemble pruning can be advantageous over using only ensemble accuracy, and that diversity measures can act as regularizers in this context

2

On new methods of dynamic ensemble selection based on randomized reference classifier

Krysmann M., Kurzyński M.

Journal of Medical Informatics & Technologies

|

2012

|

Vol. 20

101--107

EN

In the paper two dynamic ensemble selection (DES) systems are proposed. Both systems are based on a probabilistic model and utilize the concept of Randomized Reference Classifier (RRC) to determine the competence function of base classifiers. In the first system in the selection procedure of base classifiers the dynamic threshold of competence is applied. In the second DES system, selected classifiers are combined using weighted majority voting rule with continuous-valued outputs, where the weights are equal to the class-dependent competences. The performance of proposed MCSs were tested and compared against DES system with better-than-random selection rule using eleven databases taken from the UCI Machine Learning Repository. The experimental results clearly show the effectiveness of the proposed methods.

3

Classification techniques for non-invasive recognition of liver fibrosis stage

Krawczyk B., Woźniak M., Orczyk T., Porwik P., Musialik J., Błońska-Fajfrowska B.

Journal of Medical Informatics & Technologies

|

2012

|

Vol. 20

121--127

EN

Contemporary medicine should provide high quality diagnostic services while at the same time remaining as comfortable as possible for a patient. Therefore novel non-invasive disease recognition methods are becoming one of the key issues in the health services domain. Analysis of data from such examinations opens an interdisciplinary bridge between the medical research and artificial intelligence. The paper presents application of machine learning techniques to biomedical data coming from indirect examination method of the liver fibrosis stage. Presented approach is based on a common set of non-invasive blood test results. The performance of four different compound machine learning algorithms, namely Bagging, Boosting, Random Forest and Random Subspaces, is examined and grid search method is used to find the best setting of their parameters. Extensive experimental investigations, carried out on a dataset collected by authors, show that automatic methods achieve a satisfactory level of the fibrosis level recognition and may be used as a real-time medical decision support system for this task.

4

Speaker recognition based on the combination of GMM and SVDD

Zhou Y., Zhang X., Wang J., Gong Y., Zhou Y.

Przegląd Elektrotechniczny

|

2011

|

R. 87, nr 3

329-332

EN

Scare-level combination of subsystems can yield significant performance gains over individual subsystems in speaker recognition. A novel speaker verification method based on support vector data description (SVDD) is proposed to remedy the defect of Gaussian mixture model (GMM) to same extent, and then using the theory of multiple classifier systems (MCS),a new speaker recognition system based on the combination of GMM and SVDD is proposed. Experiments on TlMIT speech database show that the GMM-SVDD model fully utilizes the complementarities of GMM and SVDD to improve the performance obviously in speaker verification, closed-set speaker identification and speaker recognition.

PL

Zaproponowano nowa metodę rozpoznawania głosu bazującą na systemie SVDD jako alternatywę dla modelu GMM. Następnie wykorzystując teorię wielokrotnego systemu klasyfikacji MCS zaproponowano wykorzystanie połączenia systemów GMM i SVDD. Eksperymenty potwierdziły że nowy model GMM-SVOO umożliwia ulepszonę rozpoznawanie głosu.

5

Cost-sensitive classifier ensemble for medical decision support system

Woźniak M., Zmyślony M.

Journal of Medical Informatics & Technologies

|

2011

|

Vol. 17

97--104

EN

Multiple classifier systems are currently the focus of intense research. In this conceptual approach, the main effort focuses on establishing decision on the basis of a set of individual classifiers' outputs. This approach is well known but usually most of propositions do not take exploitation cost of such a classifier under consideration. The paper deals with the problem how to take a test acquisition cost during classification task under the framework of combined approach on board. The problem is known as cost-sensitive classification and it has been usually considered for the decision tree induction. In this work we adapt mentioned above idea into choosing members of classifier ensemble and propose a method of choosing a pool of individual classifiers which take into consideration on the one hand quality of ensemble on the other hand cost of classification. Properties of mentioned concept are established during computer experiments conducted on chosen medical benchmark databases from UCI Machine Learning Repository.

6

Combining classifiers - concept and applications

Woźniak M.

Journal of Medical Informatics & Technologies

|

2010

|

Vol. 15

19--27

EN

Problem of pattern recognition is accompanying our whole life, therefore methods of automatic pattern recognition is one of the main trend in Artificial Intelligence. Multiple classifier systems (MCSs) are currently the focus of intense research. In this conceptual approach, the main effort is concentrated on combining knowledge of the set of individual classifiers. Proposed work presents a brief survey of the main issues connected with MCSs and provides comparative analysis of some classifier fusion methods.

7

A Robust Multiple Classifier System for Pixel Classification of Remote Sensing Images

Maulik U., Chakraborty D.

Fundamenta Informaticae

|

2010

|

Vol. 101, nr 4

286-304

EN

Satellite image classification is a complex process that may be affected by many factors. This article addresses the problem of pixel classification of satellite images by a robust multiple classifier system that combines k-NN, support vector machine (SVM) and incremental learning algorithm (IL). The effectiveness of this combination is investigated for satellite imagery which usually have overlapping class boundaries. These classifiers are initially designed using a small set of labeled points. Combination of these algorithms has been done based on majority voting rule. The effectiveness of the proposed technique is first demonstrated for a numeric remote sensing data described in terms of feature vectors and then identifying different land cover regions in remote sensing imagery. Experimental results on numeric data as well as two remote sensing data show that employing combination of classifiers can effectively increase the accuracy label. Comparison is made with each of these single classifiers in terms of kappa value, accuracy, cluster quality indices and visual quality of the classified images.

8

Feature Selection Algorithm for Multiple Classifier Systems: A Hybrid Approach

Delimata P., Suraj Z.

Fundamenta Informaticae

|

2008

|

Vol. 85, nr 1-4

97-110

EN

Many problems in pattern classification and knowledge discovery require a selection of a subset of attributes or features to represent the patterns to be classified. The approach presented in this paper is designed mostly for multiple classifier systems with homogeneous (identical) classifiers. Such systems require many different subsets of the data set. The problem of finding the best subsets of a given feature set is of exponential complexity. The main aim of this paper is to present ways to improve RBFS algorithm which is a feature selection algorithm. RBFS algorithm is computationally quite complex because it uses all decision-relative reducts of a given decision table. In order to increase its speed, we propose a new algorithm called ARS algorithm. The task of this algorithm is to decrease the number of the decision-relative reducts for a decision table. Experiments have shown that ARS has greatly improved the execution time of the RBFS algorithm. A small loss on the classification accuracy of the multiple classifier used on the subset created by this algorithm has also been observed. To improve classification accuracy the simplified version of the bagging algorithm has been applied. Algorithms have been tested on some benchmarks.

9

A Rough Set Approach to Multiple Classifier Systems

Suraj Z., El Gayar N., Delimata P.

Fundamenta Informaticae

|

2006

|

Vol. 72, nr 1-3

393-406

EN

During the past decade methods of multiple classifier systems have been developed as a practical and effective solution for a variety of challenging applications. A wide number of techniques and methodologies for combining classifiers have been proposed in the past years in literature. In our work we present a new approach to multiple classifier systems using rough sets to construct classifier ensembles. Rough set methods provide us with various useful techniques of data classification. In the paper, we also present a method of reduction of the data set with the use of multiple classifiers. Reduction of the data set is performed on attributes and allows to decrease the number of conditional attributes in the decision table. Our method helps to decrease the number of conditional attributes of the data with a small loss on classification accuracy.