Sparse learning of interval type-2 fuzzy model with embedded feature and rule selection

A new approach of GA-based type reduction of interval type-2 fuzzy model for nonlinear MIMO system: Application in methane oxidation process

Reliable knowledge of project cash flow is essential for effective project management. Since the nature of project is uncertain and easily effected by different criteria, considering uncertainty has become a vital part of any effective project management approach. Cash, as one of the most important project resources, requires reliable management techniques. In this paper, a new interval type-2 fuzzy project cash flow analysis model based on interval type-2 fuzzy project scheduling is proposed to predict project cash flow in different periods of project life cycle. Despite using type-2 fuzzy sets in various practical problems, they are new to project management; thus, they are used in this paper to address the uncertainty and lack of sufficient knowledge in project activity durations and costs. In the first part, interval type-2 fuzzy project scheduling model is introduced, and then the forward and the backward passes are defined. In order to improve this part, a new flexible and controllable method of interval type-2 fuzzy ranking is introduced. In the second part, a model of cash flow assessment is proposed under an interval type-2 fuzzy environment. This model is able to calculate the minimum and the maximum cash distribution, cash flow of project in different periods and total direct cost of projects in addition to cash flow uncertainty. The presented model provides the manager with a comprehensive insight into the cash required in different stages of project and helps the managers to locate the periods with highest uncertainty. The model also uses the concept of alpha-levels to analyze different levels of risk of uncertainty. For the purpose of illustration, the proposed interval type-2 fuzzy model is used to generate cash flow of main activities of a real construction project in a developing country, and the results are presented. The results present high flexibility of model in expressing and considering uncertainty and vagueness, in addition to its high capability in the risk evaluation. Finally, the results of the case study are used to present the managerial implications.

Image segmentation is a popular application area of machine learning. In this context, each target region drawn from an image is defined as a class towards recognition of instances that belong to this region (class). In order to train classifiers that recognize the target region to which an instance belongs, it is important to extract and select features relevant to the region. In traditional machine learning, all features extracted from different regions are simply used together to form a single feature set for training classifiers, and feature selection is usually designed to evaluate the capability of each feature or feature subset in discriminating one class from other classes. However, it is possible that some features are only relevant to one class but irrelevant to all the other classes. From this point of view, it is necessary to undertake feature selection for each specific class, i.e, a relevant feature subset is selected for each specific class. In this paper, we propose the so-called multi-task feature selection approach for identifying features relevant to each target region towards effective image segmentation. This way of feature selection requires to transform a multi-class classification task into $n$ binary classification tasks, where $n$ is the number of classes. In particular, the Prism algorithm is used to produce a set of rules for class specific feature selection and the K nearest neighbour algorithm is used for training a classifier on a feature subset selected for each class. The experimental results show that the multi-task feature selection approach leads to an significant improvement of classification performance comparing with traditional feature selection approaches.

Compliance monitoring for quality of web service (QoWS) has accuracy issues due to uncertain network behaviors. Existing models use precise computation-based methods for defining and monitoring QoWS requirements, but these methods have limited ability to handle uncertainties. Consequently, the accuracy of the monitoring results is degraded. Defining expected QoWS using exact values is unrealistic, as generally not all service requestors know what values should be specified in the contract. Therefore, this paper proposes an interval type-2 (IT2) fuzzy model for QoWS compliance monitoring because it has greater capability than precise computation methods to reduce the effects of uncertainties. IT2 also has greater capability than the traditional fuzzy sets to manage uncertainty problem due to its non-crisp membership degrees assigned to the input. The model is able to perform compliance monitoring on linguistically defined QoWS. The model is developed based on fuzzy C-means algorithm, and the number of clusters is optimized using a clustering validity index. The model is constructed based on a Mamdani fuzzy inference system. The results show that the IT2 model outperforms type-1 fuzzy and precise computation-based models in terms of the accuracy of monitoring results. This research results in more accurate and precise QoWS compliance monitoring. It also provides user-centric QoWS specifications because requestors can define their requirements using linguistic values.

Results are presented of testing different feature sets and classification rules for Slovene phone recognition purposes. Different parametric descriptions of speech such as Fourier, cepstral, and linear prediction analysis of 12.8-ms long overlapping signal frames were used. It was found that linear filter spacing, gives significantly better recognition results for some phone categories than logarithmic spacing. Recognition results using LPC-cepstrum parametric representation were not significantly different from those obtained by MEL-cepstrum when using the Bayes classifier assuming a normal distribution of the feature vector. >

To solve the disaster problem caused by high-dimensional data, this paper proposes a new feature selection algorithm that combines kernel functions with sparse learning. Firstly, each dimension feature of the data set is mapped to a high-dimensional kernel space by using a kernel function, and each dimension of the data set performs linear feature selection in the high-dimensional kernel space to achieve nonlinear feature selection in low-dimensional space. Secondly, we sparsely reconstruct the features mapped to the kernel space to get a sparse representation of the original data set. Meanwhile, we use l 1 -norm to construct feature scoring selection mechanism and select the optimal feature subset. Finally, this data subset is used for classification experiments. Experimental results on public datasets showed that the proposed algorithm can conduct feature selection better, and the accuracy of classification can be increased by about 4%.

In this work, we propose a novel Feature Selection framework, called Sparse-Modeling Based Approach for Class Specific Feature Selection (SMBA-CSFS), that simultaneously exploits the idea of Sparse Modeling and Class-Specific Feature Selection. Feature selection plays a key role in several fields (e.g., computational biology), making it possible to treat models with fewer variables which, in turn, are easier to explain, by providing valuable insights on the importance of their role, and might speed the experimental validation up. Unfortunately, also corroborated by the no free lunch theorems, none of the approaches in literature is the most apt to detect the optimal feature subset for building a final model, thus it still represents a challenge. The proposed feature selection procedure conceives a two steps approach: (a) a sparse modeling-based learning technique is first used to find the best subset of features, for each class of a training set; (b) the discovered feature subsets are then fed to a class-specific feature selection scheme, in order to assess the effectiveness of the selected features in classification tasks. To this end, an ensemble of classifiers is built, where each classifier is trained on its own feature subset discovered in the previous phase, and a proper decision rule is adopted to compute the ensemble responses. In order to evaluate the performance of the proposed method, extensive experiments have been performed on publicly available datasets, in particular belonging to the computational biology field where feature selection is indispensable: the acute lymphoblastic leukemia and acute myeloid leukemia, the human carcinomas, the human lung carcinomas, the diffuse large B-cell lymphoma, and the malignant glioma. SMBA-CSFS is able to identify/retrieve the most representative features that maximize the classification accuracy. With top 20 and 80 features, SMBA-CSFS exhibits a promising performance when compared to its competitors from literature, on all considered datasets, especially those with a higher number of features. Experiments show that the proposed approach might outperform the state-of-the-art methods when the number of features is high. For this reason, the introduced approach proposes itself for selection and classification of data with a large number of features and classes.

Producción Científica

Knowledge discovery from genomic data has become an important research area for biologists. Nowadays, a lot of data is available on the web, but the corresponding knowledge is not necessarly also available. For example, the first draft of the human genome, which contains 3×109 letters, has been achieved in June 2000, but up to now only a small part of the hidden knowledge has been discovered. The aim of bioinformatics is to bring together biology, computer science, mathematics, statistics and information theory to analyze biological data for interpretation and prediction. Hence many problems encountered while studying genomic data may be modeled as data mining tasks, such as feature selection, classification, clustering, and association rule discovery. An important characteristic of genomic applications is the large amount of data to analyze and it is, most of the time, not possible to enumerate all the possibilities. Therefore, we propose to model these knowledge discovery tasks as combinatorial optimization tasks, in order to apply efficient optimization algorithms to extract knowledge from large datasets. To design an efficient optimization algorithm, several aspects have to be considered. The main one is the choice of the type of resolution method according to the characteristics of the problem. Is it an easy problem, for which a polynomial algorithm may be found? If yes, let us design such an algorithm. Unfortunately, most of the time the response to the question is ‘NO’ and only heuristics, that may find good but not necessarily optimal solutions, can be used. In our approach we focus on evolutionary computation, which has already shown an interesting ability to solve highly complex combinatorial problems. In this chapter, we will show the efficiency of such an approach while describing the main steps required to solve data mining problems from genomics with evolutionary algorithms. We will illustrate these steps with a real problem.

A Feature Selection and Association Rule Approach to Identify Genes Associated with Metastasis and Low Survival in Sarcoma

Data quality on categorical attribute is a difficult problem that has not receivedas much attention as numerical counterpart. Our basic idea is to employ association rule for the purpose of data quality measurement. Strong rule generation is an important area of data mining.Association rule mining problems can be considered as a multiobjective problem rather than as a single objective one.The main area of concentration was the rules generated by association rule mining using gene tic algorithm. The advantage of using geneticalgorithm is to discover high level predictionrules is that they perform a global search and cope better with attribute interaction than the greedy rule induction algorithm often used in data mining. Genetic algorithm based approach utilizes the linkage betweenassociation rule and feature selection. In this paper, we put forward a Multi objective genetic algorithm approach for data quality on categorical attributes. The result shows that our approach is outperformed by the objecti ves like accuracy, completeness, comprehensibilityand interestingness.

Chaotic systems are nonlinear dynamic systems, the main feature of which is high sensitivity to initial conditions. To initiate a design process in fuzzy model, chaotic systems must first be represented by T-S fuzzy models. In this paper, a new fuzzy modeling method based on sector nonlinearity approach has been recommended for chaotic systems relating to initial condition variations using the interval type-2 Takagi–Sugeno (IT2 T-S) fuzzy model. Examining many famous chaotic systems, it can be seen that nonlinear terms in chaotic systems are composed of just one variable or more. In the process of constructing an IT2 T-S fuzzy model which represents the chaotic systems, authors will focus on nonlinear terms of the chaotic systems. The proposed interval type-2 Takagi-Sugeno fuzzy modeling method is employed for two kinds of nonlinear terms; at first, a uni-variable nonlinear term is presented and then a multi-variable one will be introduced. So, it will be shown how many famous chaotic systems are represented by IT2 T-S fuzzy model. Then the proposed approach is applied to Genesio-Tesi and Rossler systems. Numerical simulations are given to demonstrate the efficiency of the proposed method in MATLAB environment.

Exclusive feature selection and multi-view learning for Alzheimer’s Disease

Feature selection has recently been the subject of intensive research in data mining, especially for datasets with a large number of descriptive attributes such as feature selection in customer relationship management (CRM). In this paper, FRI algorithm which has some deficiencies in feature selection of market segments groups is improved. A new FC-based GMDH model is built. It has the advantage of combining both qualitative and quantitative information in the decision analysis, which is extremely important for CRM. To derive the decision rules from different customer group for identifying features that contribute to CRM, both fuzzy clustering and heuristic algorithm are developed in this paper. It has been demonstrated in the empirical research that the proposed algorithm is able to derive the rules and identify the most significant features more accuracy than FRI when feature difference between customer segments is not obvious, which is unique and useful in solving CRM problems. The results showed the practical viability of the proposed approach for customer feature selection.

Simultaneous learning of decision rules and important attributes for classification problems in image analysis