Data cleaning

1. Data collection

The data is obtained through web scraping and comes in raw GeoJSON format, often including unnecessary or redundant information.

2. Data Cleaning

The raw data data gets cleaned:

Character normalization: Special Portuguese characters are standardized to simplify further processing.
Polygon correction: GeoJSON polygons are cleaned to fix inconsistencies.

3. Data Transformation

the data is transformed into a structured SQLite database using csv_to_sqlite. This format allows for efficient data access, quick querying, and seamless integration into the platform.

Data prediction

Crime prediction using raster grid from QGIS

Each raster cell contains crime data and variable for the prediction

QGIS Steps:

Create fishnet for Belem
Load Sector from PA_Setores2021.zip
OSM Bus data
Join attributes by Intersect Fishnet
Transform to raster
Make prediction
- Raster Training file

KNN and Random Forest Classifier outperformed the Logistic Regression with high precision suggesting fewer false alarms.

Logistic Regression

Using

y = {
    1, # Cell with accident,
    0 # Cell without accident
}

LRM is used o to measure the relationships between the dependent variable and thepredictors. The logistic regression model’s output is a predicted transformed logistic (logit)probability map with continuous values ranging from 0 to 1

The inspiration is from a well structured study analyzing traffic accident in iran.

References

[1] (Pedestrian Road Traffic Accidents in Metropolitan Areas: GIS-Based Prediction Modelling of Cases in Mashhad, Iran)(https://www.researchgate.net/publication/372109711_Pedestrian_Road_Traffic_Accidents_in_Metropolitan_Areas_GIS-Based_Prediction_Modelling_of_Cases_in_Mashhad_Iran)

QGIS

The geojsondata is loaded with qgis, a powerful tool to visualize, analyze geospatial data. Sectors data and a fishnet help to divide the problem to smaller section to then make a prediction.

A classiification output the most probable crime in each sector givin insight in how to resolve the crime problem.

Python analysis

EDA

Data exploration is done with to have a first look at the data to find patterns and to check if the data has something to tell to us like area that form clusters indicading populated area with high crimes, dangerous streets or crime organizations.

Data analysis

With an accuracy of 83% given also the high amount of data spannig 3 years is successful predicting the type of crime given the crime variables used.

A web platform for navigating crime data, allowing users to explore what crimes occurred, where they happened, and when they took place. Users can filter and combine these parameters to view the data in a table or visualize patterns and relationships through interactive charts.

Crimes Dashboard