MIDS W207 Final Project

Kaggle San Francisco Crime Classification Competition

https://www.kaggle.com/c/sf-crime

Initial Features

• Category -Category of the crime incident. This is the target variable for prediction
• Dates -Timestamp of the crime incident
• DayOfWeek -The day of the week
• PdDistrict -Name of the Police Department District
• Address -The approximate street address of the crime incident
• X -Longitude
• Y -Latitude

Engineered Features

• 'X' and 'Y' coordinates were trimmed to four digits of significance and paired to allow larger ‘plots’ of land to be accurately related
• Holidays were identified using the Pandas built-in calendar for US holidays
• Dates feature was broken into the following: -- Hour of the Day -- Day of the week (dummy variable) -- Week of the year

Matplotlib to visualize parameter optimization and identify knees

Baseline Model

K-NN

Our baseline model for this analysis was a K-NN model with K=50.

Additional Models

Neural Network

Optimizations and methods:

• Keras (on top of TensorFlow) for NNs
• Sigmoid activation and ReLU based models
• Modeled with and without dropouts

Random Forests

Final Parameters:

Optimizations and methods:

• Boosting
• Bagging
• GridSearchCV for parameter optimization