Comparative Analysis of Traditional Machine Learning Methods and Spiking Neural Networks for Spatio-temporal Data Mining
A new framework in this study, which uses spiking neural networks for learning spectro-temporal and spatio-temporal data, is the NeuCube. The NeuCube is able to learn and classify and predict data, both in online and offline modes. NeuCube-based methodology is used, tested and implemented for the classification and regression of spatio-temporal data (seismic data set). The spatio-temporal data consists of both time and space. In this study the spatial data are latitude, longitude and depth. The temporal data is the magnitude of the earthquake. The modelling of the spatio-temporal data is used to predict new patterns from the complex spatio- and spectral temporal data and to make an accurate prediction of events such as the predicting the occurrence of earthquakes. Seismic data is in relation to the occurrence of earthquakes and is acquired by analyzing the surface of the earth through the deployment of various sensors. After deployment, actuation of the sensor’s source, which receives the seismic signals, occurs in turn producing raw seismic data. In this study, the author performs a comparative analysis of the spatio-temporal data with regards to the machine learning algorithms (WEKA), evolving connectionist systems (NeuCom) and spiking neural networks (NeuCube). The comparative analysis between machine learning algorithms and spiking neural networks is based on classification of the data, and the comparative analysis between evolving connectionist systems and spiking neural networks is based on regression/prediction. The seismic dataset used in this study is publicly available on the GeoNet website (www.geonet.org.nz).