This is a project to model the New York City 311 Hotline complaint type based on time and users' location.
The New York City 311 Hotline data were downloaded from NYC Open Data and were served on AWS RedShift at the time of this project
There are 23.5 MM rows and 41 columns in this dataset in total. I sub-selected 500k rows from 2014-2019 to perform this analysis.
- Developed a multi-class classification LightGBM model to predict the top 10 complaint types by count from 2014-2019.
- Utilized the top-3 accuracy as metrics due to the nature of this problem (considered to be correct as long as the top-three predictions match).
- Built a data processing pipeline for data cleaning, missing value imputation (KNN), and categorical variable encoding.
- Created both time-lag and geo- features to correlate time and location to the target.
- Achieved a top-3 accuracy of 0.97, which suggests that the caller's complaint type is correlated with their locations and time of the call.
- Visualized the complaint types spatially via Tableau, and the dashboard is available upon request.
- Python 3.6.5
- Matplotlib 2.2.2
- Seaborn 0.9.0
- Pandas 0.23.0
- Scikit-Learn 0.19.1
- LightGBM 2.2.1