Machine Learning has been widely used for making earthquake predictions due to their ability to improve over time. With the huge amount of earthquake instrumental data, machine learning approaches are capable enough to improve efficiency and accuracy in earthquake prediction. [3] Multiple machine learning methods including, Artificial Neural Network (ANN), Support Vector machine (SVM), K-nearest neighbour (KNN), Naive Bayes (NB) and random forest algorithms have been exercised for earthquake prediction.
Many studies have been made on the damage rate of buildings in Japan. Mononobe [1970] was the first to investigate them the overturning of tomb stones which is equivalent to ground seismic coefficient shows the relations obtained from the damage data of several large earthquakes in Japan and theoretical curves expressed by normal probability distribution function. [4] Shiga [1976] discussed the damage rate of concrete buildings those which are reinforced based on his research on the earthquake resistance capacity of existing those which are reinforced buildings. He derived a simple formula to estimate roughly the resistance capacity of lower reinforced concrete buildings using the amount of column areas and wall areas through the analysis of damaged as well as undamaged buildings in the (1968) Tokachioki earthquake and the (1978) Miyagikenoki earthquake. He estimated the probability distribution of the earthquake resistance capacity of existing low-rise reinforced concrete buildings and modeled the probability distribution by Gamma distribution, based on which he gave the relation between the damage rate of reinforced concrete buildings and the ground motion intensity.
Wenrui Li, Nakshatra, Nishita Narvekar, Nitisha Raut, Birsen Sirkeci, Jerry Gao introduce us to the idea that a strong earthquake is followed by aftershocks. We can detect location of these aftershocks by analysis of arrival time of P-waves and S-waves. Data collection from 16 earthquake stations in SAC file format, which contains time series data and is a waveform, used by authors to study trends in P-waveand S-wave. Data is clipped followed by noise removal to only obtain needed waveform by means of triggering algorithm and filters. [5] AR picker algorithm used to determine values of P-wave and S-wave arrival time which are treated as extracted feature. Waveform is then converted into ASCII format. Data is then fed to different machine learning models-SVM, Decision trees Random forest and linear regression for comparison purpose. Random Forest distinguishes between earthquake leading and non-earthquake leading data the best, with an accuracy of 90. Use of triangulation technique to calculate epicenter, predict arrival time of P-wave and S-wave and the difference between the two arrivals.
Earthquake activity is presumed as a spontaneous phenomenon that can damage huge number of lives and properties, and currently there is no any model exists that can predict the exact position, magnitude, frequency and time of an earthquake. Researchers have conducted several experiments on earthquake events and forecasts, leading to a variety of findings based on the factors considered. [6] The well-known Gutenberg and Richter statistical model found a correlation between the magnitude of earthquake and frequency of earthquake. For structural design, this earthquake probability distribution model was used. In supervision of the California Geological Survey, Petersen conducted research and suggested a model that is time-independent. This time independent model demonstrating that chances of occurrence of earthquake follow the Poisson's distribution model. Shen suggested a probabilistic earthquake forecasting model based on the strain studied between the behaviour of tectonic plates. Based on this model, higher measured strain results in a higher risk of earthquake.