A crime prediction tool based on Heterogeneous Clustering and an original evaluation metric

Table of Contents

• Introduction
• Getting Started

Introduction

This tool implements a crime prediction algorithm in a geological space using heterogeneous clustering and an evaluation metric. The algorithm and the metric is proposed by the Data Science Lab at USC (http://dslab.usc.edu/).

Getting Started

1. Install Python

2. Install Jupyter Notebook (only if you want to visualize the results)

3. Install all the packages listed in packages.txt. If you have pip installed, you could run pip install -r packages.txt to install all the packages.

4. Edit config.py for parameter settings:

   ignoreFirst - int: Minimum amount of training periods

   periodsAhead_list - List of ints: Periods ahead to forecast

   ug_gridshapes - List of tuples: # of cells along latitude & longitude (for uniform grid method)

   ug_maxDist - Leave at 0 (for uniform grid method)

   ug_threshold - Leave at 0 (for uniform grid method)

   ug_methods - List of str: Any of ["mm", "ar", "harmonic]. Forecasting algorithms to use (for uniform grid method)

   c_gridshape - Tuple: # of cells along latitude & longitude (for cluster method)

   c_thresholds - int: Threshold of clustering (for cluster method)

   c_maxDist - int: Neighborhood distance of clustering (for cluster method)

   c_methods - List of str: Any of ["mm", "ar", "harmonic]. Forecasting algorithms to use (for cluster method)

   resource_indexes - List of int: List of amount of resources to use for evaluation (RA calculation)

   cell_coverage_units - int: Number of resources needed to cover each cell (RA calculation)

5. Sample usage for forecasting & evaluation (using LAdata.pkl):

   python parse_data.py DPSdata.pkl DPSUSC.pkl
   python make_predictions.py LAdata.pkl
   python calculate_resource_allocation.py
   python plot_allocations.py