We implemented Markov-modulated Poisson process into antenna-antenna communication in Senegal between 1 Jan 2013 – 31 May 2013.

Our work is based on an assumption that people tend to make calls, when extraordinary conditions take place. Each antenna’s hourly call volume reproduces a characteristic feature, depending on location, and time. This temporal pattern, high during the day and low at night, resembles a heart electrocardiogram. Our work aims to learn this temporal pattern as normal behaviour of each antenna and predict the possible irregularities as anomalous events. These events may represent crime incidents, public demonstrations, natural disasters, which are crucial for public safety and security. Identifying the severity and location of the event and immediate actions may save lives, especially in developing countries where source of information may not be very reliable.

ACLEDEvents We use ACLED event database to evaluate our methodology.

First we analyse the antenna’s call volume per time. Here you can find the antenna’s call observation, and hourly mean’s and deviation from that mean of SenegalZiguinchor region.



A sample video from D4D Senegal event detection work.