Wyniki wyszukiwania - BazTech

1

Bandwidth selection for kernel generalized regression neural networks in identification of hammerstein systems

Lv Jiaqing, Pawlak Mirosław

Journal of Artificial Intelligence and Soft Computing Research

|

2021

|

Vol. 11, No. 3

181--194

EN

This paper addresses the issue of data-driven smoothing parameter (bandwidth) selection in the context of nonparametric system identification of dynamic systems. In particular, we examine the identification problem of the block-oriented Hammerstein cascade system. A class of kernel-type Generalized Regression Neural Networks (GRNN) is employed as the identification algorithm. The statistical accuracy of the kernel GRNN estimate is critically influenced by the choice of the bandwidth. Given the need of data-driven bandwidth specification we propose several automatic selection methods that are compared by means of simulation studies. Our experiments reveal that the method referred to as the partitioned cross-validation algorithm can be recommended as the practical procedure for the bandwidth choice for the kernel GRNN estimate in terms of its statistical accuracy and implementation aspects.

2

Soft computing-based technique as a predictive tool to estimate blast-induced ground vibration

Arthur Clement Kweku, Temeng Victor Amoako, Ziggah Yao Yevenyo

Journal of Sustainable Mining

|

2019

|

Vol. 18, iss. 4

287--296

EN

The safety of workers, the environment and the communities surrounding a mine are primary concerns for the mining industry. Therefore, implementing a blast-induced ground vibration monitoring system to monitor the vibrations emitted due to blasting operations is a logical approach that addresses these concerns. Empirical and soft computing models have been proposed to estimate blast-induced ground vibrations. This paper tests the efficiency of the Wavelet Neural Network (WNN). The motive is to ascertain whether the WNN can be used as an alternative to other widely used techniques. For the purpose of comparison, four empirical techniques (the Indian Standard, the United State Bureau of Mines, Ambrasey-Hendron, and Langefors and Kilhstrom) and four standard artificial neural networks of backpropagation (BPNN), radial basis (RBFNN), generalised regression (GRNN) and the group method of data handling (GMDH) were employed. According to the results obtained from the testing dataset, the WNN with a single hidden layer and three wavelons produced highly satisfactory and comparable results to the benchmark methods of BPNN and RBFNN. This was revealed in the statistical results where the tested WNN had minor deviations of approximately 0.0024 mm/s, 0.0035 mm/s, 0.0043 mm/s, 0.0099 and 0.0168 from the best performing model of BPNN when statistical indicators of Mean Absolute Error (MAE), Root Mean Square Error (RMSE), Relative Root Mean Square Error (RRMSE), Correlation Coefficient (R) and Coefficient of determination (R2) were considered.

3

Automated detection of the preseizure state in EEG signal using neural networks

Sudalaimani C., Sivakumaran N., Elizabeth Thomas T., Rominus Valsalam S.

Biocybernetics and Biomedical Engineering

|

2019

|

Vol. 39, no. 1

160--175

EN

The life-threatening neural syndrome epilepsy is elicited by seizure which affects over 50 million people in the universal. A seizure is a brain condition made by excessive, unusual exoneration by nerve cells of the brain. Contemporary seizure forecast research works exhibited worthy results in both undersized and lengthy electroencephalography (EEG) signal; however it is essential to formulate superior epileptic seizure forecast system; that shall be steady, constant and less resource intensive for effectively employed to heading for evolving a convenient and easily manageable ictal or seizure forewarning prearrangement or devices. Based on our exploration, we have found a novel seizure prediction method which we evaluated by producing ten sub-frequency EEG data from initially recorded signal. Simple, robust and computationally less-intense EEG characteristics are mined using the generated sub-frequency signals and applied the extracted features to computationally less intense generalized regression neural network (GRNN) to segregate EEG signal clips into normal or preseizure files. In this research work, we have engendered 10 sub-frequency bands of signals from original EEG recordings, extracted various meaningful features from those sub-frequency band signals, created 10 GRNN neural networks to categorize feature files as normal or preseizure, and then applied post-processing techniques with 10 thresholding mechanisms to each classifier output. As such, we determined that seizure fore-warning may function better in various sub-frequency bands for many patients in a subject-specific manner. We also found that epileptic-seizure forecast performed superior at '60 Hz high pass' filtered sub-frequency band EEG signal for all subjects or canines data.

4

Pattern layer reduction for a generalized regression neural network by using a self-organizing map

Kartal S., Oral M., Ozyildirim B. M.

International Journal of Applied Mathematics and Computer Science

|

2018

|

Vol. 28, no. 2

411--424

EN

In a general regression neural network (GRNN), the number of neurons in the pattern layer is proportional to the number of training samples in the dataset. The use of a GRNN in applications that have relatively large datasets becomes troublesome due to the architecture and speed required. The great number of neurons in the pattern layer requires a substantial increase in memory usage and causes a substantial decrease in calculation speed. Therefore, there is a strong need for pattern layer size reduction. In this study, a self-organizing map (SOM) structure is introduced as a pre-processor for the GRNN. First, an SOM is generated for the training dataset. Second, each training record is labelled with the most similar map unit. Lastly, when a new test record is applied to the network, the most similar map units are detected, and the training data that have the same labels as the detected units are fed into the network instead of the entire training dataset. This scheme enables a considerable reduction in the pattern layer size. The proposed hybrid model was evaluated by using fifteen benchmark test functions and eight different UCI datasets. According to the simulation results, the proposed model significantly simplifies the GRNN’s structure without any performance loss.

5

An epileptic seizure detection system based on cepstral analysis and generalized regression neural network

Yavuz E., Kasapbaşı M. C., Eyüpoğlu C., Yazıcı R.

Biocybernetics and Biomedical Engineering

|

2018

|

Vol. 38, no. 2

201--216

EN

This study introduces a new and effective epileptic seizure detection system based on cepstral analysis utilizing generalized regression neural network for classifying electroen-cephalogram (EEG) recordings. The EEG recordings are obtained from an open database which has been widely studied with many different combinations of feature extraction and classification techniques. Cepstral analysis technique is mainly used for speech recognition, seismological problems, mechanical part tests, etc. Utility of cepstral analysis based features in EEG signal classification is explored in the paper. In the proposed study, mel frequency cepstral coefficients (MFCCs) are computed in the feature extraction stage and used in neural network based classification stage. MFCCs are calculated based on a frequency analysis depending on filter bank of approximately critical bandwidths. The experimental results have shown that the proposed method is superior to most of the previous studies using the same dataset in classification accuracy, sensitivity and specificity. This achieved success is the result of applying cepstral analysis technique to extract features. The system is promising to be used in real time seizure detection systems as the neural network adopted in the proposed method is inherently of non-iterative nature.