Chaotic time-series prediction with artificial neural networks: The case of earthquake data

The purpose of this thesis is how successful chaotic time-series can be predicted after theprogress has been made in chaos analysis and artificial intelligence. Since the accuraterecordings began in 1971 magnitudes of the earthquakes occured in Marmara region between1971 and 2005 are utilized to generate the time-series. It is thought that because of thedeterministic, dynamic, nonlinear, complex structure of chaos which follows a definite orderfits also in earthquakes? features, thus it is appropriate for this thesis? outline. As an addition,earthquakes which have been tried to be predicted for centuries and have caused hugecasualties makes these types of studies meaningful and necessary.In order to acquire this goal there are used some signal processing methods, analysis toolspresented after chaos theory, artificial neural networks and some programs for both analysisand prediction.There has been acquired significant success rates from the prediction of the time-series whichare chaotic and long enough with minimum level of anomaly. Short-term prediction resultshas been better than long-term prediction?s. However, success in prediction of the generaltrend has not reflected to the prediction of major earthquakes.Keywords: Chaotic Time-series Prediction, Earthquake Prediction, Chaos Theory, ArtificialNeural Networks, Time-series Analysis.