Narrow Results

Approximately, one third of patients with epilepsy are refractory to medical therapy and thus can be at high risk of injuries and sudden unexpected death. A low-complexity electroencephalography (EEG)-based seizure monitoring algorithm is critically important for daily use, especially for wearable monitoring platforms. This paper presents a personalized EEG feature selection approach, which is the key to achieve a reliable seizure monitoring with a low computational cost. We advocate a two-step, personalized feature selection strategy to enhance monitoring performances for each patient. In the first step, linear discriminant analysis (LDA) is applied to find a few seizure-indicative channels. Then in the second step, least absolute shrinkage and selection operator (LASSO) method is employed to select a discriminative subset of both frequency and time domain features (spectral powers and entropy). A personalization strategy is further customized to find the best settings (number of channels and features) that yield the highest classification scores for each subject. Experimental results of analyzing 23 subjects in CHB-MIT database are quite promising. We have achieved an average F-1 score of 88% with excellent sensitivity and specificity using not more than 7 features extracted from at most 3 channels.

A new approach for nonlinear system identification based on Takagi-Sugeno fuzzy models is presented. The premise structure and membership functions are optimized by the LOLIMOT (local linear model tree) algorithm, see [1]. This method is extended by a subset selection technique which automatically determines the structure of the local linear models in the rule consequents. This allows to select the significant input variables for static models and additionally the determination of the dynamic orders and dead times for dynamic models. The utilized subset selection technique is the orthogonal least-squares (OLS) algorithm. It exploits the linear regression structure of the problem and thus is very fast. The applicability of the proposed approach is illustrated by the identification of a transport delay process which has operating point dependent time constants and dead times.

The main goal of software quality engineering is to produce a high-quality software product through the use of various techniques and processes. Classification models are effective tools for software quality prediction, helping practitioners to detect potentially problematic modules and eventually improve software product. However, two potential problems, high dimensionality and class imbalance, may affect the classifiers performance. In this study, we propose a data pre-processing approach, in which feature selection is combined with data sampling, to overcome these problems. We investigate two filter-based feature subsets selection techniques, i.e., correlation-based and consistency-based subset evaluation methods, and three data sampling methods, i.e., random undersampling, random oversampling, and synthetic minority oversampling. We are interested in exploring the effect of the various feature selection techniques, sampling methods, and their interactions on the performance of classification models. The empirical studies were carried out on 13 datasets from two real-world software systems. The results demonstrate that the correlation-based subset evaluation technique outperformed the consistency-based method when they were used along with a random sampling method and when the training data had a high degree of class imbalance; however, when synthetic minority oversampling was employed or when the training dataset was less imbalanced, the consistency-based technique had better performance than the correlation-based approach.

In time series modeling, one problem is to identify a small number of influential factors to explain variations in the variable of interest. With a vast number of possible factors available, suitable features need to be identified to yield multi-factor models with good explanatory power. In this paper, we propose a novel subset selection method which makes use of the properties in the frequency domain environment. The proposed system ensures key patterns in the target variable be sought and suitable factors be selected based on frequency peaks in common. It can perform well even when the number of factors is significantly greater than the sample size. Moreover, a very important feature of the proposed system is the capability of handling factors with different timeframes, which is lacking in existing methods. We demonstrate the system via several examples with dataset from finance, economic, road traffic and air pollution.

Selecting a system or a subset of systems that have the best performance among many alternative systems is an important area of research. All efforts are paid to do this selection with a high probability of correct selection $P(CS)$. In this paper, we compare the performance of two popular procedures used to select the best systems, namely, the method of Kim and Nelson $({𝒦}{𝒩})$ which is considered as a ranking and selection procedure, and the Optimal Computing Budget Allocation $({𝒪}{𝒞}{ℬ}{𝒜})$ procedure. Both of these procedures determine the number of samples for each system that are needed to achieve the maximum $P(CS)$. Recently, the ${𝒪}{𝒞}{ℬ}{𝒜}$ has been developed in order to be able to approximate the value of $P(CS)$, so both procedures approximate the value of $P(CS)$. To do the comparison, we implement the two methods on two examples; one is a generic example and the other one involves a management of inventory model. The two examples assure that the ${𝒪}{𝒞}{ℬ}{𝒜}$ method always outperforms the ${𝒦}{𝒩}$ method.

Most enterprise datasets are large, but some are very small for predictive purposes due to expensive experiments, reduced budget or tight schedule required to generate them. The bootstrap approach is a method used frequently for small datasets in data mining. Numerous theoretical studies have been done on bootstrap in the past two decades but few have applied it to solve real world manufacturing problems. Bootstrap methods provide an attractive option when model selection becomes complex due to small sample sizes and unknown distributions. In principle, bootstrap methods are more widely applicable than the jackknife method, and also more dependable. In this chapter we focus on selecting the best model based on prediction errors computed using the revised bootstrap method, known as the 0.632 bootstrap. Models developed and selected are then clustered and the best cluster of models is next bagged to provide the minimum prediction errors. Numerical examples based on a small enterprise dataset illustrate how to use this procedure in selecting, validating, clustering, and bagging predictive regression models when sample sizes are small compared to the number of parameters in the model.