Contributors: Andrew Deng, Amir Kazi, Tianchu Shu and Jessica Song

Syntax Error aims to identify which inmates are most at risk of recidivism following their release within one year or two year, using personal data, mental health records from Johnson county jail system and demographic data of neighborhood areas

• Data exploration
• Data cleaning and preprocessing
• Feature generation
• Machine learning pipeline
• Model evaluation and bias analysis

1. pandas
2. psycopg2
3. numpy
4. matplotlib
5. seaborn
6. graphviz
7. sklearn
8. datetime
9. requests

Johnson County Jail, Census Bureau

We have worked on

1. Data Exploration
2. Data Cleaning, Data Integration, and Pre-Processing
3. Feature Generation
4. Machine learning Classifiers & Evaluations (Pipeline)

1. Jupyter notebook

• Person_societal_var.ipynb
• Bail_info.ipynb
• mh_info.ipynb
• All-var.ipynb

2. Python code ("/code")

• census.py
• run_api.py

• final_default_grids.py
• jocojims.py
• indpv_lists.py

• final_connection.py
• final_load_dfs.py
• final_explore_and_viz.py
• final_preprocessing.py
• final_temporal.py
• final_classifier_final.py
• final_plot.py
• final_pipeline.py

• final_run.py

• Via Jupyter Notebook

• Via Python file & Terminal

  python final_run.py 
  

In general, all work for preliminary steps of the project is located in raw， all code for the pipeline can be found in code with the prefix 'final' (eg. final_plot).

We compared the effectiveness of different models trained on “biased” and “unbiased” feature sets:

1. All variables (Using all our variables for training data)
2. Personal & Mental Health related Variables 11
3. Personal & Bail related Variables
4. Personal related Variables.