Winda Astuti

Hybrid Technique Using Singular Value Decomposition (SVD) and Support Vector Machine (SVM) Approach for Earthquake Prediction

Most of the existing earthquake (EQ) prediction techniques involve a combination of signal processing and geophysics techniques which are relatively complex in computation for analysis of the Earths electric field data. This paper proposes a relatively simpler and faster method that involves only signal processing procedures. The prediction of the EQ occurrence estimation using a combination of singular value decomposition (SVD)-based technique for feature extraction and support vector machine (SVM) classifier is presented in this paper. Using the proposed method, the Earths electric field signal is transformed into a new domain using SVD-based approach. In this approach, the time domain signal is projected on the left eigenvectors of impulse response matrix of the linear prediction coefficient (LPC) filter. Several features have been extracted from the transformed signal. These features are used as input for the SVM classifier in order to predict the location of the forthcoming EQ. Once the location is determined, a similar approach is used to estimate its magnitude. Finally, the time estimation of the forthcoming EQ is estimated based on the statistical observation. The occurred EQs during 2008 in Greece are used to train the classifiers, whereas those occurred from 2003 to 2010 in the same region are used to evaluate the performance of the proposed system. In predicting the location of the future EQs, the proposed system could achieve 77% accuracy. As for the magnitude prediction, the proposed system provides an accuracy of 66.67%. Moreover, the predicted time for the EQ with magnitude greater than Ms = 5 is 2 days ahead, whereas for magnitude greater than Ms = 6 is up to 7 days ahead.

Adaptive Short Time Fourier Transform (STFT) Analysis of seismic electric signal (SES): A comparison of Hamming and rectangular window

Seismic electric signal (SES) is one of features for predicting earthquakes (EQs) because of its significant changes in the amplitude of the signal prior to the earthquake. This paper presents detailed analysis of SES recorded prior to earthquake that occurred in Greece in the period from January 1, 2008 to June 30, 2008. During this period of time 5 earthquakes were recorded with magnitudes greater than 6R. In this analysis STFT involving adaptively sliding window technique is used, in which Hamming and rectangular window functions are applied and compared. The comparison shows that Hamming window gives better results in analyzing the first significantly changes of SES prior to the EQ. The application of Hamming window resulted in less rippled spectrum shape which is more suitable to be used in characterizing the SES.

Animal sound activity detection using multi-class support vector machines

On March 11th 2011, the whole world was taken aback by another tragic experience of Tsunami triggered by a magnitude 9.8 earthquake in Japan. Just few days after that, on March 25th 2011, another earthquake of magnitude 6.8 hit Myanmar deaths and destructions. Despite the loss incurred on properties and human being, available data show that relatively few numbers of animals died during most natural disasters. Prior to the occurrence of these disasters, available reports shows that animals do migrate to higher level or leave the areas en masse ahead of the event. Other related account show that animal sometimes behaves in unusual ways prior to the occurrence of these natural disasters. These overwhelming evidences point to the fact that animals might have the ability to sense impending natural disaster precursor signals ahead of time. This paper discusses the preliminary results obtained from the use of support vector machine (SVM) and Mel-frequency cepstral coefficients (MFCC) in the development of animal sound activity detection (ASAD) which is an integral part in the development of earthquake and natural disaster prediction using unusual animal behavior. The use of MFCC has been proposed for the features extraction stage while SVM has been proposed for classification of the extracted features. Preliminary results obtained shows that the MFCC and SVM can be used for features extraction and features classification respectively.

Graphical based monitoring of the Earth's electric field signal prior to the earthquake

Earthquake is one of the most destructive natural disasters that kill many people and destroy a lot of properties. Therefore, it is highly important to know well ahead the occurrence of earthquakes in order to reduce the number of victims and material losses. Earthquake prediction is one solution to reduce such disastrous effect. The Earths electric field signal is one of potential features that can be used for earthquake prediction generated from the released energy through a sudden dislocation of the segment in the earths crust, this signal has become one of the potential features to predict the earthquake, since it has significant changes in the amplitude of the signal prior to the earthquake. The significant changes of the Earths electric field signals prior to the earthquake are detected and utilized in this work. This paper presents the development of the Graphical based monitoring system for the earths electric field. The system uses a graphical user interface (GUI) developed using MATLAB ®. The work is illustrated on earthquake data which occurred in Greece, between January 1, 2008 and June 30, 2008. In that period of time, thirteen earthquakes had occurred. Six of them were recorded with magnitudes greater than Ms=5R (5R), while seven of them were recorded with magnitudes greater than Ms=6R (6R).

Singular Value Decomposition (SVD) Based Orthogonal Transform Approach for Earth's Electric Field Signal Processing

The Earths electric field signal is generated from the released energy through a sudden dislocation of the segment in the earths crust. Many researchers have reported the use of parametric modeling technique for earths electric field signal processing. The existing earths electric signal processing based on parametric modeling technique has suffered from the noise. Therefore, the effective earths electric field signal processing is necessary in order to process the signal with better performance for the identification. Singular value decomposition (SVD) based parametric modeling technique is applied as feature extraction technique to the Earths electric field signal. The projection of excitation signal on the right eigenvector of the LPC filter impulse response matrix is involved in this technique. The combination of SVD-based parametric modeling technique has perfectly classified the significant Earths electric field data prior to the earthquake and the Earths electric field data on the normal condition after the polynomial kernel function is applied.

Time domain feature extraction technique for earth’s electric field signal prior to the earthquake

Earthquake is one of the most destructive of natural disasters that killed many people and destroyed a lot of properties. By considering these catastrophic effects, it is highly important of knowing ahead of earthquakes in order to reduce the number of victims and material losses. Earths electric field is one of the features that can be used to predict earthquakes (EQs), since it has significant changes in the amplitude of the signal prior to the earthquake. This paper presents a detailed analysis of the earths electric field due to earthquakes which occurred in Greece, between January 1, 2008 and June 30, 2008. In that period of time, 13 earthquakes had occurred. 6 of them were recorded with magnitudes greater than Ms=5R (5R), while 7 of them were recorded with magnitudes greater than Ms=6R (6R). Time domain feature extraction technique is applied to analyze the 1st significant changes in the earths electric field prior to the earthquake. Two different time domain feature extraction techniques are applied in this work, namely Simple Square Integral (SSI) and Root Mean Square (RMS). The 1st significant change of the earths electric field signal in each of monitoring sites is extracted using those two techniques. The feature extraction result can be used as input parameter for an earthquake prediction system.