What Dataset am I going to use ?

Ample data streams were required to hypothesize and validate any anomaly detection algorithm. Fortunately, it has been a while since CSN set up its network of inexpensive sensors. Hence CSN servers had ample data stored for the task, which spanned across 400 hundred sensors and 36 events.

Some of them are:

Earthquakes in Southern California

Name	UTC Date and Time
La Canada	17-10-2013 12:35
Joshua Tree	06-10-2013 02:06
La Verne 2	19-09-2013 12:06
La Verne	19-09-2013 11:43
Weldon 2	25-08-2013 18:50
Weldon	24-07-2013 16:46
Rancho Palos Verdes	07-06-2013 11:25
Santa Barbara Channel	29-05-2013 14:38
Meiners Oaks	07-05-2013 09:06
Marina del Rey 2	29-04-2013 03:06
Marina del Rey	27-04-2013 02:52
Los Angeles	20-03-2013 16:22
Fontana 2	17-03-2013 14:00
Anza 2	13-03-2013 04:21
Anza	11-03-2013 16:56
Rancho Cucamonga	10-02-2013 20:26
Avalon	14-12-2012 10:36
North Hollywood 2	19-11-2012 15:44
Monterey Park	14-11-2012 00:25
Lennox	05-11-2012 04:06
Manhattan Beach	06-11-2012 02:39
San Fernando	08-11-2012 03:33
Newhall 2	28-10-2012 15:24
Beverly Hills	03-09-2012 10:26
Beverly Hills Long	03-09-2012 10:26
Brawley	26-08-2012 20:57
Yorba Linda 2	08-08-2012 16:33
Altadena	05-07-2012 18:48
Altadena Long	05-07-2012 18:48
Fontana 3	15-01-2014 09:35
Fontana Monica	15-01-2014 09:35
Fontana 3 Long	15-01-2014 09:35
Fontana 3 Long Millikan	15-01-2014 09:35
Hollywood	08-02-2013 18:13
Westwood	17-03-2014 13:25
La Habra	29-03-2014 04:09

A CSN Sensor

The task remaining was to collect this data from the servers. CSN has a command line interface which allows data download in very convenient data SAC format, which ObsPY can read and handle operations on it very well. I download over 1200 streams of data, each of which had following characteristics:

• Each stream represents one channel of one sensor (one axis of acceleration X, Y and Z)
• Each stream was exactly 20 minutes long and had acceleration values the collected from the sensor by the server.
• The sampling rate of the data was 50 samples per second.
• An earthquake occurred in each of data stream at precisely 18th minute, on the basis time declared by United States Geological Survey (USGS) official report.

Thus a typical data stream looked like the one below:

Figure 2. Plot generated from ObsPy look like this. This is data stream where each point is acceleration value at time instant the sensor experiences.