Feature Selection Using Artificial Neural Networks

Ledesma, Sergio; Cerda, G.; Aviña, Gabriel; Hernández, Donato; Torres, Miguel

doi:10.1007/978-3-540-88636-5_34

Cited by 15 publications

(9 citation statements)

References 7 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Both spatial and spectral resolutions of the sensors have increased; thus, the selection of the most appropriate data as inputs, known as feature selection, has become a more critical issue, particularly for neural networks. For this purpose, the pruning of neural networks has been suggested as an alternative to existing statistical methods [12][13][14][15].…”

Section: Introductionmentioning

confidence: 99%

Design of Feedforward Neural Networks in the Classification of Hyperspectral Imagery Using Superstructural Optimization

2020

View full text Add to dashboard Cite

Artificial Neural Networks (ANNs) have been used in a wide range of applications for complex datasets with their flexible mathematical architecture. The flexibility is favored by the introduction of a higher number of connections and variables, in general. However, over-parameterization of the ANN equations and the existence of redundant input variables usually result in poor test performance. This paper proposes a superstructure-based mixed-integer nonlinear programming method for optimal structural design including neuron number selection, pruning, and input selection for multilayer perceptron (MLP) ANNs. In addition, this method uses statistical measures such as the parameter covariance matrix in order to increase the test performance while permitting reduced training performance. The suggested approach was implemented on two public hyperspectral datasets (with 10% and 50% sampling ratios), namely Indian Pines and Pavia University, for the classification problem. The test results revealed promising performances compared to the standard fully connected neural networks in terms of the estimated overall and individual class accuracies. With the application of the proposed superstructural optimization, fully connected networks were pruned by over 60% in terms of the total number of connections, resulting in an increase of 4% for the 10% sampling ratio and a 1% decrease for the 50% sampling ratio. Moreover, over 20% of the spectral bands in the Indian Pines data and 30% in the Pavia University data were found statistically insignificant, and they were thus removed from the MLP networks. As a result, the proposed method was found effective in optimizing the architectural design with high generalization capabilities, particularly for fewer numbers of samples. The analysis of the eliminated spectral bands revealed that the proposed algorithm mostly removed the bands adjacent to the pre-eliminated noisy bands and highly correlated bands carrying similar information.

show abstract

Section: Introductionmentioning

confidence: 99%

Design of Feedforward Neural Networks in the Classification of Hyperspectral Imagery Using Superstructural Optimization

2020

View full text Add to dashboard Cite

show abstract

“…When used with a large number of features and layers, these models are difficult to interpret as the estimated parameters of the model, known as weights, are not directly convertible to a meaningful measure of relevance. However, given our simple design, the MLP can be used effectively for feature selection (59). To determine the importance of each feature, the dependent resampled input method was used (60).…”

Section: Top Featuresmentioning

confidence: 99%

Data-Driven Discovery of Predictors of Virtual Reality Safety and Sense of Presence for Children With Autism Spectrum Disorder: A Pilot Study

et al. 2020

View full text Add to dashboard Cite

Virtual reality (VR) offers children with autism spectrum disorder (ASD) an inexpensive and motivating medium to learn and practice skills in a personalized, controlled, and safe setting; however, outcomes of VR interventions can vary widely. In particular, there is a need to understand the predictors of VR experience in children with ASD to inform the design of these interventions. To address this gap, a sample of children with ASD (n=35, mean age: 13.0 ± 2.6 years; 10 female) participated in a pilot study involving an immersive VR experience delivered through a head-mounted display. A data-driven approach was used to discover predictors of VR safety and sense of presence among a range of demographic and phenotypic user characteristics. Our results suggest that IQ may be a key predictor of VR sense of presence and that anxiety may modify the association between IQ and sense of presence. In particular, in low-anxiety participants, IQ was linearly related to experienced spatial presence and engagement, whereas, in high-anxiety participants, this association followed a quadratic form. The results of this pilot study, when replicated in larger samples, will inform the design of future studies on VR interventions for children with ASD.

show abstract

“…Speeding up a data mining algorithm, improving the data quality, improving the performance of data mining and increasing the clarity of the mining results are the significance of feature selection methods. The key benefits of feature selections are [2,3,6]:- Reduce Overfitting  Improves Accuracy  Reduce Training Time…”

Section: Feature Selectionmentioning

confidence: 99%

Exploring the High Potential Factors that Affects Students’ Academic Performance

Kaviyarasi¹,

Balasubramanian²

2018

IJEME

View full text Add to dashboard Cite

The rapid increase in student population has resulted in expansion of educational facilities at all level. Nowadays, responsibilities of teachers are many. It is the responsibilities of teachers to guide the students to choose their carrier field according to their abilities and aptitudes. The Data Mining field mines the educational data from large volumes of data to improve the quality of educational processes. Today's need of educational system is to develop the individual to enhance problem solving and decision making skills in addition to build their social skills. Educational Data Mining is one of the applications of Data Mining to find out the hidden patterns and knowledge in Educational Institutions. There are three important groups of students have been identified: Fast Learners, Average Learners, and Slow Learners. In fact, students are probably struggles in many factors. This work focuses on finding the high potential factors that affects the performance of college students. This finding will improve the students' academic performance positively.

show abstract

Feature Selection Using Artificial Neural Networks

Cited by 15 publications

References 7 publications

Design of Feedforward Neural Networks in the Classification of Hyperspectral Imagery Using Superstructural Optimization

Design of Feedforward Neural Networks in the Classification of Hyperspectral Imagery Using Superstructural Optimization

Data-Driven Discovery of Predictors of Virtual Reality Safety and Sense of Presence for Children With Autism Spectrum Disorder: A Pilot Study

Exploring the High Potential Factors that Affects Students’ Academic Performance

Contact Info

Product

Resources

About