Boston Crime Data Project

Project Description

This project was created to examine crime data from the City of Boston. Specifically, this project focuses on examining crime data collected by the City of Boston and examining trends in fraud crime occurences at several periods. The onset of the COVID pandemic and it's impact is seen through the visualization of this data, which leads to further questions about crime data forecasting before and after the pandemic, both within Fraud crimes and crime rates in general

Goals

Create deliverables:

• READ ME
• Final Report
• Functional wrangle.py, explore.py, and model.py files
• Acquire data from Analyze Boston Website
• Prepare and split the data
• Explore the data and define hypotheses, running appropriate statistical tests to accept or reject each null hypothesis
• Fit and train three (3) time-series models to predict the amount of fraud crimes on validation data
• Evaluate the models by comparing their performance validation data utilizing RMSE score
• Select the best model and evaluate it on test data
• Develop and document findings, takeaways, recommendations and next steps

Initial Thoughts

• Boston Fraud Crime will increase post-pandemic

The Plan

• Aquire and clean the data

• Explore data in search of time series patterns for fraud crimes

  • Answer the following questions
    • What was the average fraud crime rate in the two years immediately prior to the pandemic?
    • Are certain months more likely to demonstrate greater reported fraud crimes?
    • Do Fraud Crimes follow a seasonal pattern?
    • Is the downtrend seen in 2017 significant?
    • Are pre-pandemic Fraud crime levels equal to Fraud crime levels post-pandemic?

• Develop a Model to predict fraud crimes

  • Evaluate models on train and validate data
  • Select the best model to use on test data

• Draw Conclusions

Data Dictionary

Feature	Definition
incident_number	Unique code representing a crime incident
offense_code	Numerical offense code unique to Boston PD
offense_code_group	Categorization of offense code by crime type
offense_description	Description of Offense including subcategory
district	Boston PD Substation of Responsibility for crime occurence location
reporting_area	numerical code for geographic area of city
shooting	Whether or not crime is a shooting
occured_on_date	Date of crime occurence
year	Year the crime occured
month	Month the crime occured
day_of_week	Day of week the crime occured
hour	Hour the crime occured
ucr_part	Uniform Crime Report Part based on crime type
street	Street name crime occured on
lat	Latitude of crime location
long	Longitude of crime location
location	Latitude and Longitude of crime location

Steps to Reproduce

1. Clone this repository
2. Put the data in the file containing the cloned repo
3. Utilize code from project notebook to aggregate data into one .CSV, which can be used locally
4. Run notebook

Takeaways and Conclusions

• Our model for predicting fraud crimes using a simple average was fairly effective
• Fraud crimes spiked during the pandemic
• This visualization shows that since lockdown restrictions were eased in Boston in the spring of 2021, Fraud crimes appears to be decreasing
• When aggregating fraud crimes by day, there was low seasonality in this data
• Pre-pandemic, a simple average model was the best model for predicting fraud crime
• A simple average model performed extremely poorly on test data, which was the first year of COVID. In Boston, COVID restrictions meant strict lockdowns for the early stages of the pandemic
• Though lockdown restrictions have been completely lifted, the fraud crime rate has significantly decreased in Boston

Recommendations

• Resampling fraud crimes by week or month to get a clearer picture of seasonality
• Creating a new model for post-pandemic data, excluding lockdown periods due to severe outliers

Next Steps

• Literature Review to examine trends observed in this project compared to other examinations of crime data during the pandemic
• Examining Boston Crime Data for other crimes besides fraud, to see if other types of crime are more seasonal
• Examining Boston Crime Data for other crimes besides fraud, to see if COVID lockdowns created similar patterns of anomalies (spikes/dips) in and around lockdown periods
• Using location data to visualize crime types around the city (for a more Boston-centric approach)