Büyük verilerin zaman seri analiz metotlarının karşılaştırılması

Time series analysis is interpretation of data recorded at regular or irregular time intervals. The analysis is applied to earthquakes, sales, financial investment forecasting, early warning, decision making and many other areas. There are quantitative and qualitative methods of time series analysis of big data. The aim of this study is to compare two important quantitative methods: Autoregressive Integrated Moving Average and Long Short-Term Memory. For comparison, the accuracy of earthquake magnitude and location prediction has been used. Earthquake data (earthquake magnitude, longitude, latitude; sun and moon altitude-azimuth-distance to earth at the instant of the earthquake) between years 1970 and 2019 were utilized. Eighty percent of the United States Geological Survey data was used for training and the remaining for testing. Earthquakes of magnitude 4.0 and above were considered. The results of each method were compared with the results of previous works, using standard deviation, mean squared error, mean absolute error and median absolute error performances. ARIMA was inefficient in long term predictions. Therefore, LSTM was deemed as the appropriate method for the analysis of long term, irregular data, as supported in the literature too. Vanilla and Stacked models of LSTM were compared and Stacked LSTM provided the best results after being optimized by changing the dense, batch size and epoch parameters. The best results were obtained by using parameters 8 dense, 64 batch size and 64 epochs. While the longitude prediction was way off, the best predicted outcomes were the magnitude and latitude of the earthquakes.