Lazaro Martull

Dallas Crime Data: Predicting Crime Locations Using Machine Learning

Overview

This project builds machine learning models to predict crime incident locations (premise) using Dallas Police open data. The task is a multi-class classification problem where the target variable is the reported location type and the inputs are the crime type and category.

Data

• Source: Dallas Open Data Portal (Socrata API)
• Time Range: 2014–2024
• Initial Pull: 10,000 records via API
• Final Clean Dataset: 9,790 records after cleaning and removing unclear values

Key fields (renamed):

• servyr → year
• offincident → crime
• nibrs_crime_category → crime_category
• premise → location (target variable)

Methodology

• Removed missing or unclear values (notably in nibrs_crime_category)
• One-hot encoded categorical features
• Trained models on 2014–2023 data
• Evaluated performance on a 2024 holdout set
• Used 3-fold cross-validation during training
• Compared results for:
  • All location classes
  • Top 5 most frequent locations (to reduce class imbalance)

Models

• Logistic Regression
• K-Nearest Neighbors (KNN)
• Random Forest

Results (Highlights)

All Locations (2024 Holdout)

Holdout Accuracy:

• Logistic Regression: 0.0667
• Random Forest: 0.0400
• KNN: 0.1867

AUC:

• Logistic Regression: 0.7175
• Random Forest: 0.7054
• KNN: 0.6107

Top 5 Locations (2024 Holdout)

Filtering to the top five most common location classes significantly improved performance:

Holdout Accuracy:

• Logistic Regression: 0.4211
• Random Forest: 0.4211
• KNN: 0.2895

Limitations & Future Work

• Performance decreases when predicting across many location classes due to multi-class imbalance and problem complexity
• Potential improvements include:
  • Collecting additional data
  • Hyperparameter tuning
  • Exploring advanced models such as SVMs, gradient boosting, or deep learning

Files

• 📄 Final Report: report.pdf
• 🧪 Notebook / Code: analysis.ipynb