Modelling burglary susceptibility - an expert judgement approach

Abstract

Developing a model allow a better understanding of the nature of a complicated phenomenon. With advancement of tools and technology, model development has been applied widely to mimic the phenomena of interest, spatial or non-spatial wise, allowing a guided decision making to be made. In this paper, the phenomena of burglary vulnerability and susceptibility are modelled based on expert opinion input to create a model that imitates the expert profiling of burglary occurrences, which is dependent on individual expert wisdom and experience in handling the burglary investigation. Due to seriousness of burglary crime offences in Malaysia, especially the urban areas, a prediction model is needed to correlates the factor of crime and further estimates the spatial susceptibility to work hand in hand with other government initiatives in reducing crime. Eighteen (18) indicators and 63 sub-indicators has been identified to be significant in defining the susceptibility of burglary. Apart from input of rating and ranking of indicators and sub-indicators obtained from questionnaire distribution to expert in handling burglary, the geospatial based data were also incorporated into the model to add the element of spatial accuracy in susceptibility prediction. The geospatial data includes the distribution of burglary incidence from 2010 - 2016, the census data, the building footprint data and the demarcation area. For the collected questionnaire feedback, the procedure of Analytical Hierarchy Process (AHP) were adapted to determine the weight value considering the rating input of expert from the distributed questionnaire. The input of weight and scoring were applied to the corresponding spatial features and combined with the operation of weighted sum to yield the total burglary susceptibility of a place. The results of the model were validated with the real reported burglary frequency based on True Positive Rate correlation matrix. The model validation finds that the model have a sensitivity of 82% in classifying the burglary susceptibility of the building polygon inside the study area. However this model still requires some improvement as it is still lacking to perform the classification of incidence intensity correctly.