Week 07 (W51 Dec28) Crimes in the UK - Rostlab/DM_CS_WS_2016-17 GitHub Wiki

Besides conducting descriptive statistics that we will continue and publish next week, we will present in this Wiki entry the prediction goals of our project.

We will use the following data sets for building the models used for the prediction goals:

  • Data.police.uk data (crimes and stop and search)
  • POI
  • Demographics
  • Police forces

For each prediction goal we provide a short description, an expected input (which can differ from the input used for building the model), an expected output, and finally possible applications, only if these are not already obvious.

  1. Effects of demographic structure on criminal behaviour: We aim to predict the effects of demographic structure on criminal behaviour. Our goal is to build a model that will be trained to predict the number of the crimes broken down by the different crimes types for a certain area given its demographic data.
    Input: Demographic data of a certain area [, month (January, …)]
    Output1: Number of crimes in the given month (total, crime_type1, crime_type2,…)
    Output2: Number of crimes between September 2011 and September 2016 (total, robbery, …)
    Applications:
  • Predict the criminal behaviour of areas in the UK, where we don’t have any police data.
  • We believe/hope that this model can be applied to other similar countries.

  1. Anomaly detection: Given the demographic and criminal data about the different areas in the UK, we aim to conduct a predictive anomaly detection. With its help we shall detect abnormal areas, with respect to criminal behaviour that deviates remarkably from the other similar areas.
    Input: -
    Output: Abnormal areas.

  2. Stop and Search Probability for a person that visits an area: Our data set comprises information about stop and searches (S&S) in the UK since 2014. Unlike the crime data, S&S provide more background information about the suspects (age, ethnicity, gender,...) and the police officer(s) who handled the case. With the help of this data we aim to predict the probability of a person getting stopped and searched in a certain area, given his/her background.
    Input: Information about the person, demographic data of the destination area, a week day
    Output: The probability of getting stopped and searched in this areas in the given week day.
    Application:

  • Generate a map with the probabilities of a specific person getting stopped and searched in different areas in the UK for comparison
  1. Qualitative assessment of police forces: This prediction focuses on providing a probability of a specific crime type (in a specific area and month of the year) getting solved
    Input: crime type, month, demographic data of a certain area.
    Output: probability of a successful case clearance.
    Applications:
  • Generate a map with the probabilities of a crime getting solved in different areas in the UK.
  • Given our knowledge about the police forces responsible for the different areas in the UK, we can find out which police forces is more successful in resolving specific crime types.