Highlights
- Processed 10,000 records from 86 raw attributes
- Engineered 11 features through temporal binning and spatial clustering
- K-Means clustering identified 418 outliers (4.2% of dataset)
- Classification accuracy of 79.6% with Random Forest
- Detected anomalies with response times 5-18x above average
- Followed CRISP-DM methodology with 3-person team
About This Project
A machine learning project that analyzes Dallas Police incident records to classify crime types, identify spatial-temporal patterns, and detect anomalous incidents using classification, clustering, and outlier detection techniques.
The project processed 10,000 police incident records from 86 raw attributes, engineering 11 features through temporal binning, spatial clustering, and categorical consolidation. K-Means clustering (K=10) was applied using the Elbow Method to segment incidents, identifying 418 outliers (4.2% of dataset) across 3 anomalous clusters.
The analysis detected operational anomalies with response times 5-18x higher than the dataset average and data quality issues including impossible reporting delays. Classification models achieved 79.6% accuracy using Decision Trees and Random Forest algorithms.