Predicting Crime Categories Using Machine Learning

Project Overview

This project focuses on crime classification using machine learning models. The dataset comprises 20,000 records and 20 columns, covering aspects such as incident location, victim demographics, date, and time. The target variable includes six crime categories:

• Property Crimes
• Fraud & White Collar Crimes
• Violent Crimes
• (Additional three categories)

This analysis has significant implications for public safety, urban management, and policy-making when integrated thoughtfully.

Data Exploration, Cleaning, and Preprocessing

• Conducted thorough data exploration to understand patterns and missing values.
• Applied cleaning techniques to remove inconsistencies and outliers.
• Created 23 relevant feature columns to enhance model performance.

Machine Learning Models

Two models were utilized for classification:

• LGBMClassifier
• XGBoostClassifier

A VotingClassifier was built using these two classifiers, leading to an improved accuracy of 0.88080, significantly higher than the baseline accuracy of approximately 0.58.

Conclusion

The project demonstrates the power of machine learning in crime classification. The high accuracy of the model highlights its potential for real-world applications in crime prediction, policy formulation, and urban safety enhancements.