Introduced by Kapilesh Ramesh, Richard Man, Rijul Mediratta, Siddhant Issar, and Xianpeng Wang
Earthquakes are some of the most devastating natural disasters known to mankind today. They are caused by the faults on earth’s continuously moving tectonic plates. The earthquakes result in a release of stored energy in the form of waves which, depending on the intensity of the earthquake, can cause violent vibrations and loss of human lives. SkyAlert is using a sensor network called REDSSA (SkyAlert Seismic Detection Network) located at various locations in Mexico to detect these waves and send early earthquake warnings to users up to 120 seconds in advance so that people can evacuate and the loss of human lives can be minimized.
Our team worked with two types of data provided to us by SkyAlert. The first was individual earthquake data for 5 earthquakes that occurred in 2017 and 2018. These had 29 features and the number of rows ranging from 95 – 530. The other data set we worked on was a complete sensor log of SkyAlert from 2013 till early 2018 with 126,000 rows. The features in these data sets were of various types – date/time, numerical and categorical. Our team used all the softwares at our disposal to work with the data sets – RStudio, Python, and Excel.
Our team conducted three main types of exploratory data analysis pertaining to visualization and understanding the data. The data set used here was an 8.2 magnitude earthquake that happened in September 2017 in Mexico.
The data set consisted of a feature which had the time stamp of how various sensors in the network were activated in response to the earthquake. These were used in combination with folium and geopandas python libraries to animate the sequence of sensor activation. This video provides crucial insights on where the earthquake waves are coming from, thus giving valuable information on where the epicenter might be located.
In a nutshell, our team developed an easy to use data analytics tool for SkyAlert that will help them to analyze their system and spot anomalies in an effective and efficient manner. Furthermore, we helped understand the team at SkyAlert the sequence of sensor and server date/times and the significant features contributing to sensor activation.