ics5110-crime-statistics

US Crime Analysis: Weapon Use Prediction

Welcome to the Weapon Use Prediction project! This interactive web application leverages machine learning models to predict the type of weapon (Firearm vs. Non-Firearm) used in crimes based on demographic and contextual features. The app is built using Gradio and hosted on Hugging Face Spaces.

Latest code:

• Git repo: https://github.com/dgrixti/ics5110-crime-statistics
• Hugging Face: https://huggingface.co/spaces/grixtid/ICS5110/tree/main

• Git page: https://dgrixti.github.io/ics5110-crime-statistics/

🔍 Overview

The primary goal of this project is to analyze crime data and classify weapon usage using machine learning techniques. The application supports multiple models, including:

• Random Forest Classifier
• XGBoost
• Logistic Regression
• Artificial Neural Network (ANN)

The models are trained on a publicly available crime dataset, which includes features such as Region, Victim Age, Relationship Type, and more. The dataset is imbalanced - 102,988 records (67.09%) for “Firearm” compared to approximately 50,523 (32.91%) for “Non-Firearm”, and techniques like SMOTE were applied to improve model performance.

---

🌐 Try It Out

Hugging Face Space

🚀 Features

• Interactive Interface: Enter key features like Region, Season, Relationship Type, etc., and get predictions in real time.
• Multiple Models: Compare predictions from different machine learning algorithms.
• Performance Metrics: View Accuracy, Precision, Recall, F1-Score, and Confusion Matrix for model performance evaluation.
• Real-World Use Case: Designed for researchers and law enforcement agencies to analyze crime patterns.

---

💻 How to Use

1. Access the Application:
   • Visit the Hugging Face Space to interact with the app. The iframe below also provides direct access.
2. Input Features:
   • Select or input values for features like Region, Victim Sex, Relationship Type, etc.
   • Adjust Victim Age using the slider.
3. Choose a Model:
   • Select a model from the dropdown menu: Random Forest, XGBoost, Logistic Regression, or ANN.
4. View Predictions:
   • The app displays the predicted weapon category (Firearm or Non-Firearm) along with detailed metrics.

---

📊 Technical Details

Dataset

• Source: US Crime Dataset on Kaggle
• Features: Demographics, geographic data, relationships, and crime specifics.

Models

• Random Forest: Robust ensemble method with SMOTE for handling imbalance.
• XGBoost: Boosting algorithm with optimized hyperparameters for precision-recall balance.
• Logistic Regression: Lightweight model with decent accuracy.
• ANN: Neural network with three dense layers for binary classification.

Preprocessing

• Categorical features: Encoded using LabelEncoder.
• Numerical features: Standardized using StandardScaler.
• Class imbalance: Addressed using SMOTE for minority-class oversampling.

---