In this project, I predicted the likelihood each street segment in DC (portions of a street between two intersections or end of a street) experienced a homicide between 1 January 2007 and 31 October 2017, based on f1 score, through a boosted tree classification model using XGBoost based on neighborhood characteristics such as zoning and median property values. I further simplified the model by turning homicide incidence into a Boolean feature instead of a continuous feature and running an XGBoost classification, although an XGBoosted tree regression could easily be used on raw homicide counts per street for a more granular prediction of how many homicides a given street is predicted to have experienced. Finally, I will visualized the results on map using QGIS.
The following project requires the following packages are installed:
- Numpy
- Pandas
- Scikit-Learn
- XGBoost
- Seaborn
- Graphviz
Once the required packages are installed and updated, please execute the following steps to load the homicides environment:
$ cd ~/Predicting_DC_Homicides
$ conda env create -f homicides.yml
Once the environment is loaded, execute the following to activate the new environment:
- Windows:
$ activate homicides - macOS and Linux:
$ source activate homicides
Now, check that the environment was installed correctly by running:
$ conda list
The homicides environment should appear in the list and should have an asterix next to it in the far left.
Execute the following to launch the Jupyter Notebook, and proceed through the notebook to run the model on the data:
$ Jupyter Notebook
The repository includes:
- Environment .yml file
- Data
FINAL_DATASETin csv format - Metadata -explains the features and label in
FINAL_DATASET - Jupyter Notebook
- A PDF finalized report Predicting DC Homicides
- MIT License
- README.md file
This project is licensed under the MIT License