Efficient Feature Subset Selection and Subset Size Optimization

Somol, Petr; Novovičová, Jana; Pudil, Pavel

doi:10.5772/9356

Cited by 49 publications

(38 citation statements)

References 44 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…In the following, we will briefly present and evaluate this algorithm, along with another, general-purpose feature selection method called Sequential Forward Floating Selection (SFFS) (Somol et al, 2010). Then we will manually select a small set of filters, where the manual selection is guided by some simple heuristics.…”

Section: Gabor Filtersmentioning

confidence: 99%

“…In the following, we will briefly introduce these algorithms. As a detailed discussion of these methods is beyond the scope of our paper, we kindly refer the reader to Gramms (1991) and Somol et al (2010). We selected the size and overlap of our filters based on the literature and also on experimental findings.…”

Section: Automatic Feature Selection Algorithmsmentioning

confidence: 99%

“…A detailed analysis of this algorithm, was given by Somol et al (2010). This method consists of the following main steps:…”

Section: Sequential Forward Floating Selection (Sffs)mentioning

confidence: 99%

See 2 more Smart Citations

Selection and enhancement of Gabor filters for automatic speech recognition

Kovács

Tóth

Compernolle

2014

Int J Speech Technol

View full text Add to dashboard Cite

Abstract. Motivated by neurophysiological studies, the use of Gabor filters as acoustic feature extractors for speech recognition purposes has received increasing attention in the new millenium. As the optimal parametrization of these filters is not obvious, many researchers employ different feature selection methods to find the best filter set. In this study, however, we argue that these kinds of feature selection methods cannot fulfill this task, as we demonstrate this with results obtained from experiments. We show that one can easily construct a better filter set manually, using simple heuristic rules. Then, as an alternative to the usual filter selection methods, we propose a training method that can jointly optimize the spectro-temporal filters and the neural net acoustic model built on them. In this special neural network achitecture, the filters are incorporated into the network and employed as the lowest layer of it. This allows us to tune the filters using backpropagation, and to manipulate them directly and not through their parameters. This method also has the advantage of reducing the task of filter set enhancement to that of a simple neural net training. Next, we show that we can enhance our manually selected filter set with this novel neural net architecture using the filter coefficients as initial values for the backpropagation training. The resulting filter sets were evaluated on the phone recognition task of the TIMIT corpus, using both clean and noise contaminated data; while cross-database phone recognition performance was evaluated on the "Szeged" Hungarian broadcast news database. The results we get demonstrate that the proposed filter optimization algorithm can outperform the usual feature selection-based methods, and that the filter set obtained by fine tuning the manual filters with the neural net algorithm performs even better, beating all the other methods in terms of performance.

show abstract

Section: Gabor Filtersmentioning

confidence: 99%

Section: Automatic Feature Selection Algorithmsmentioning

confidence: 99%

See 1 more Smart Citation

Selection and enhancement of Gabor filters for automatic speech recognition

Kovács

Tóth

Compernolle

2014

Int J Speech Technol

View full text Add to dashboard Cite

show abstract

“…Nevertheless, a search does not necessarily need to be exhaustive in order for it to be optimal as demonstrated in branch and bound method and best first search approaches. However, all optimal methods can be expected to be considerably slow for high dimensional problems [3]. Thus, it is often preferable for many high dimensional problems to employ heuristic methods that compromise subset optimality for better computational efficiency.…”

Section: Introductionmentioning

confidence: 99%

“…Although the new features in the new reduced dimensional space are related to the original features, the actual interpretation of the original features and hence the relation to the original system variables is completely lost in most cases. This drawback should be taken into account when considering dimensionality reduction since the actual interpretation may be important to understand the learning process that generates the new feature space [3]. Feature extraction also often associated with computational inefficiency despite the fact that it may significantly reduce dimensional space since the new constructed features are based on transformation that involves all original features including irrelevant and redundant features.…”

Section: Introductionmentioning

confidence: 99%

A new maximum relevance-minimum multicollinearity (MRmMC) method for feature selection and ranking

Senawi

Wei

Billings

2017

Pattern Recognition

View full text Add to dashboard Cite

A substantial amount of datasets stored for various applications are often high dimensional with redundant and irrelevant features. Processing and analyzing data under such circumstances is time consuming and makes it difficult to obtain efficient predictive models. There is a strong need to carry out analyses for high dimensional data in some lower dimensions, and one approach to achieve this is through feature selection. This paper presents a new relevancy-redundancy approach, called the maximum relevance-minimum multicollinearity (MRmMC) method, for feature selection and ranking, which can overcome some shortcomings of existing criteria. In the proposed method,

show abstract

Autism spectrum disorder detection using sequential minimal optimization‐support vector machine hybrid classifier according to history of jaundice and family autism in children

Yücelbaş

2021

Concurrency and Computation

View full text Add to dashboard Cite

Summary Autism spectrum disorder (ASD) is a neurodevelopmental disorder caused by several nervous system problems that change the function of the brain. It is either congenital or occurs in the early years of life. Although the cause of ASD is not known exactly, it is generally thought to be genetic. Today, the relationship of ASD with other diseases is under investigation. One of these diseases is jaundice. It has been shown in different medical studies that the probability of children, who were jaundiced in the newborn period and/or had a family history of autism, having this disorder is higher than the others. However, studies in this area using artificial intelligence techniques are limited. For this reason, in this study, the relationship between ASD with jaundice and/or family history of autism in children was emphasized by using current machine learning techniques and analyses. Datasets were established by digitizing verbal data obtained from children between the ages of 4 and 11: (1) subjects with a family history of autism, (2) subjects with a history of jaundice, (3) subjects with both history of jaundice and family autism, and (4) subjects with no conditions. Since the verbal datasets created by the answers from the children (with or without ASD) or the people around them were converted into numerical form with 0–1 coding, mathematical operations could be performed using these datasets. The four subgroups of data mentioned above were given to the sequential minimal optimization‐support vector machine hybrid classifier after they were separated into training and test data via the stratified cross‐validation method. The results were analyzed with various statistical parameters. In addition, the sequential forward floating selection algorithm was used to determine which features were not effective for ASD detection. Obtained results from all datasets can provide a new perspective for the literature. As a result, it was determined with a 100% success rate for dataset1 and dataset3. In addition, the ASD detection rate was calculated at 95.52% in children with a history of jaundice. Finally, considerable and meaningful interpretations were made about which features according to the history of jaundice and family autism for each dataset are more effective in ASD detection in children.

show abstract

Efficient Feature Subset Selection and Subset Size Optimization

Cited by 49 publications

References 44 publications

Selection and enhancement of Gabor filters for automatic speech recognition

Selection and enhancement of Gabor filters for automatic speech recognition

A new maximum relevance-minimum multicollinearity (MRmMC) method for feature selection and ranking

Autism spectrum disorder detection using sequential minimal optimization‐support vector machine hybrid classifier according to history of jaundice and family autism in children

Contact Info

Product

Resources

About