Pasi Luukka

Feature selection using fuzzy entropy measures with similarity classifier

Feature selection plays an important role in classification for several reasons. First it can simplify the model and this way computational cost can be reduced and also when the model is taken for practical use fewer inputs are needed which means in practice that fewer measurements from new samples are needed. Second by removing insignificant features from the data set one can also make the model more transparent and more comprehensible, providing better explanation of suggested diagnosis, which is an important requirement in medical applications. Feature selection process can also reduce noise and this way enhance the classification accuracy. In this article, feature selection method based on fuzzy entropy measures is introduced and it is tested together with similarity classifier. Model was tested with four medical data sets which were, dermatology, Pima-Indian diabetes, breast cancer and Parkinsons data set. With all the four data sets, we managed to get quite good results by using fewer features that in the original data sets. Also with Parkinsons and dermatology data sets, classification accuracy was managed to enhance significantly this way. Mean classification accuracy with Parkinsons data set being 85.03% with only two features from original 22. With dermatology data set, mean accuracy of 98.28% was achieved using 29 features instead of 34 original features. Results can be considered quite good.

A classifier based on the maximal fuzzy similarity in the generalized Lukasiewicz-structure

The aim of this paper is to introduce improvements made to a classifier based on maximal fuzzy similarity. Improvements are based on the use of generalized Lukasiewicz-structure and weight optimization. The main benefits of the classifier are its computational efficiency and its strong mathematical background. It is based on many-valued logic and it provides semantic information about classification results. We show that if one chooses the power value in a right manner in the generalized Lukasiewicz-structure and the optimal weights for different feature, one can see significant enhancements in classification results.

Classification based on fuzzy robust PCA algorithms and similarity classifier

In this article, classification method is proposed where data is first preprocessed using fuzzy robust principal component analysis (FRPCA) algorithms to obtain data in a more feasible form. After this we use similarity classifier for the classification. We tested this procedure for breast cancer data and liver-disorder data. The results were quite promising and better classification accuracy was achieved than using traditional PCA and similarity classifier. Fuzzy robust principal component analysis algorithms seem to have the effect that they project these data sets in a more feasible form, and together with similarity classifier classification on accuracy of 70.25% was achieved with liver-disorder data and 98.19% accuracy was achieved with breast cancer data. Compared to the results achieved with traditional PCA and similarity classifier about 4% higher accuracy was achieved with liver-disorder data and about 0.5% higher accuracy was achieved with breast cancer data.

Similarity classifier using similarities based on modified probabilistic equivalence relations

This paper examines a classifier based on similarity measures originating from probabilistic equivalence relations with a generalized mean. Equivalences are weighted and weight optimization is carried out with differential evolution algorithms. In the classifier, a similarity measure based on the Lukasiewicz structure has previously been used, but this paper concentrates on measures which can be considered to be weighted similarity measures defined in a probabilistic framework, applied variable by variable and aggregated along the features using a generalized mean. The weights for these measures are determined using a differential evolution process. The classification accuracy with these measures are tested on different data sets. Classification results are obtained with medical data sets, and the results are compared to other classifiers, which gives quite good results. The result presented in this paper are promising, and in several cases better results were achieved.

Fuzzy similarity in multicriteria decision-making problem applied to supplier evaluation and selection in supply chain management

It is proposed to use fuzzy similarity in fuzzy decision-making approach to deal with the supplier selection problem in supply chain system. According to the concept of fuzzy TOPSIS earlier methods use closeness coefficient which is defined to determine the ranking order of all suppliers by calculating the distances to both fuzzy positive-ideal solution (FPIS) and fuzzy negative-ideal solution (FNIS) simultaneously. In this paper we propose a new method by doing the ranking using similarity. New proposed method can do ranking with less computations than original fuzzy TOPSIS. We also propose three different cases for selection of FPIS and FNIS and compare closeness coefficient criteria and fuzzy similarity criteria. Numerical example is used to demonstrate the process. Results show that the proposed model is well suited formultiple criteria decision-making for supplier selection. In this paper we also show that the evaluation of the supplier using traditional fuzzy TOPSIS depends highly on FPIS and FNIS, and one needs to select suitable fuzzy ideal solution to get reasonable evaluation.

PCA for fuzzy data and similarity classifier in building recognition system for post-operative patient data

In this article we propose a method which tackles a problem where data is linguistic instead of real valued numbers. The proposed method starts with representing data as fuzzy numbers. Then generalized principle component analysis (PCA) is used, which can be used to reduce the data dimensionality and also to clear out some irregularitises from the data. After this, the data is defuzzified and then the similarity classifier is used to get the required classification accuracy. Here post-operative patient data set is used to build this expert system to determine based on hypothermia condition, whether patients in a post-operative recovery area should be sent to Intensive Care Unit, general hospital floor or go home. What makes this task particularly difficult is that most of the measured attributes have linguistic values (e.g. stable, moderately stable, unstable, etc.). Results are compared to existing result in literature and this system provides mean classification accuracy of 62.7% where as second highest reported results are with linguistic hard C-mean with 53.3%.

A multi-expert system for ranking patents: An approach based on fuzzy pay-off distributions and a TOPSIS-AHP framework

The aim of this paper is to introduce a decision support system that ranks patents based on multiple expert evaluations. The presented approach starts with the creation of three value scenarios for each considered patent by each expert. These are used for the construction of individual fuzzy pay-off distribution functions for the patent value; a consensual fuzzy pay-off distribution is then determined starting from the individual distributions. Possibilistic moments are calculated from the consensus pay-off distribution for each patent and used in ranking them with TOPSIS. It is further showed how the analytic hierarchy process (AHP) can be used to include additional decision variables into the patent selection, thus allowing for a two-tier decision making process. The system is illustrated with a numerical example and the usability of the system and the combination of methods it includes for patent portfolio selection in the real world context is discussed.

Differential Evolution Classifier in Noisy Settings and with Interacting Variables

In this paper, we have studied the performance of a differential evolution (DE) classifier in classifying data in noisy settings. We have also studied the performance in handling extra variables which simply consists of gaussian noise. Furthermore, we have carried out the classification by adding on all two component interaction terms as extra variables into the data. Also, in this situation it is crucial to have a classifier which is tolerant to noisy variables. Namely, even though there can be interaction effects in the data that can influence classification results positively, it is usually not known a priori which particular interaction components are contributing to the classification results. Therefore, we need to add all possible combinations despite the likelihood of then creating also some noisy variables which do not influence the classification accuracy, or which actually reduce the accuracy. In experimentation, we used four widely applied test data sets; the new-thyroid, heart-statlog, Hungarian heart and lenses data sets. The results indicated the DE classifier to be robust from the noise tolerance point of view in all studied cases and situations. The results suggest that the DE classifier is useful especially in the cases where interaction effects may have a significant influence to the classification accuracy.

Similarity classifier with ordered weighted averaging operators

In this article we extend the similarity classifier to cover also ordered weighted averaging (OWA) operators. Earlier, similarity classifier was mainly used with generalized mean operator, but in this article we extend this aggregation process to cover more general OWA operators. With OWA operators we concentrate on linguistic quantifier guided aggregation where several different quantifiers are studied and on how they best suite for the similarity classifier. Our proposed method is applied to real world medical data sets which are new thyroid, hypothyroid, lymphography and hepatitis data sets. Results are very promising and show improvement compared to the earlier used generalized mean operator. In this article we will show that by using OWA operators instead of generalized mean, we can improve classification accuracy with chosen data sets.

A Classification method based on principal component analysis and differential evolution algorithm applied for prediction diagnosis from clinical EMR heart data sets

In this article we have studied the usage of a classification method based on preprocessing the data first using principal component analysis, and then using the compressed data in actual classification process which is based on differential evolution algorithm, an evolutionary optimization algorithm. This method is applied here for prediction diagnosis from clinical data sets with chief complaint of chest pain using classical Electronic Medical Record (EMR), heart data sets. For experimentation we used a set of five frequently applied benchmark data sets including Cleveland, Hungarian, Long Beach, Switzerland and Statlog data sets. These data sets are containing demographic properties, clinical symptoms, clinical findings, laboratory test results specific electrocardiography (ECG), results pertaining to angina and coronary infarction, etc. In other words, classical EMR data pertaining to the evaluation of a chest pain patient and ruling out angina and/or Coronary Artery Disease, (CAD). The prediction diagnosis results with the proposed classification approach were found promisingly accurate. For example, the Switzerland data set was classified with 94.5 % ±0.4 % accuracy. Combining all these data sets resulted in the classification accuracy of 82.0 % ±0.5 %. We compared the results of the proposed method with the corresponding results of the other methods reported in the literature that have demonstrated relatively high classification performance in solving this problem. Depending on the case, the results of the proposed method were of equal level with the best compared methods, or outperformed their classification accuracy clearly. In general, the results are suggesting that the proposed method has potential in this task.