Random forest regression is an ensemble, supervised learning algorithm capable of executing both classification and regression. Within this report, the use of the following algorithm will be implemented on an earthquake dataset which consists of all recorded occurrences of earthquakes from 1930 to 2018. Certain columns from the database will be used as target variables such as magnitude and depth to predict the following outcome based on trained data. Hyper parameter tuning will be performed to maximize the model's performance by increasing its accuracy, decreasing errors, and ensuring efficiency. The parameter in this model that contributed to the efficiency while performing hyper parameter tuning was number of estimators. Findings from the research report concluded that the model's accuracy levels were approximately 75%. Despite increasing the number of trees used, the model's accuracy did not significantly change and improve but rather significantly slowed down the run-time.
Top1. Introduction
Earthquakes are one of the most unpredictable natural disasters to occur. What can range from being a slight feeling of vibration to a catastrophe, earthquakes have many side effects and endanger humans and animals on a daily basis. This can be specifically applied to areas which are geographically located in active seismic zones. This includes countries such as Japan, Indonesia, Fiji, Tonga, Turkey, China and Iran. With that said, these countries suffered terrible catastrophes which have led to fatal damages which are still being repaired to this day (U.S. Geological Survey, 2016). Due to the lack of information on earthquakes as well as possible solutions which can aid in predicting them, a large number of individuals experience a fatal death, injuries, damages to their homes, and continue living in fear for this disaster to happen again. Not Predicting the Severity of Future Earthquakes by Employing the Random Forest Algorithm 3 only does this affect individuals on a personal level, but earthquakes significantly impact economies and cause countries/nations to spend millions and billions of dollars every year to repair damages caused by earthquakes.
According to the USGS, no scientist has been able to identify potential solutions to this issue. This is because, it isn’t possible to predict earthquakes as they do not occur based on patterns and are completely out of human control. Currently, ways in which scientists and USGS are addressing the issue is by designing a system which can detect earthquakes once it has begun (U.S. Geological Survey, 2016) .By doing so, citizens in the area affected can be notified immediately. Another step that the USGS is taking is improving the durability of buildings and structures to ensure the safety of humans and reduce/ minimize damages (american geosciences institute, 2022). Although, there have not been successful forecasts in the past, it is believed that given the numerous advantages that technology has played, machine learning can play an undeniable role in geoscience and forecasting using algorithms (Papiya Debnath, 2021).
This research report explores the idea of using a machine learning algorithm to create a model which would be able to forecast earthquakes with a moderate level of accuracy. The model chosen for this exploration is Random Forest, a model which will use regression to predict the magnitude and depth of earthquakes based on existing data provided by the dataset used.