Multiple change point detection on air pollution via genetic algorithms with bayesian-MDL on non-homogeneous Poisson periods

In this talk, the change points of the time series of PM10 of the city of Bogotá are considered.  The number of change points and their respective locations are determined using the genetic algorithm. This algorithm considers the interaction of two chromosomes (mother and father) and their mutations, to conceive new generations of descendants with suitable characteristics. In this framework, the members of a generation are the change point arrangements and their MAP values.  Nonhomogeneous Poisson processes are fitted to the number of daily exceedances of a threshold between two consecutive change points. The descendants will be qualified with a new Bayesian extension of the MDL (minimum description length) penalty. The main objective of this study is to find out the impact of the environmental public policies implemented in the city by comparing their time of implementation against the change points locations.