Empirical mode decomposition applied to Afghanistan violence data: Comparison with multiplicative seasonal decomposition

The empirical mode decomposition (EMD) is applied to violence data from Afghanistan between 2006 and 2012. Several key behaviours are identified at distinct time scales ranging from days, through weeks to months, through months to a year, and finally spanning multiple years. The identified behaviour was compared to the traditionally-used multiplicative seasonal decomposition. Unlike seasonal decomposition, the EMD does not make a-priori assumptions about periodicity, and thus was better able to identify the multi-year cycle, without the skewed trend in the vicinity of turning points of the near-annual cycle. In addition, the EMD isolated shorter time scales with distinct statistical behaviour thus enriching the opportunities for analysis of drivers at different scales. Overall, the EMD demonstrated its usefulness and applicability, enhancing the analysis of violence data. The next step is to apply it to other types of time-series in the defence context to establish it firmly as a part of the defence analysis toolbox.