Non-Emergency Response Time Prediction using Deep Learning

Jillian Magyar and Adita Kulkarni, Department of Computing Sciences, SUNY Brockport, 350 New Campus Dr, Brockport NY 14420

Resource allocation for management of non-emergency incidents is an important problem that needs to be addressed for smooth functioning of cities. In this work, we investigate how long it takes to resolve non-emergency requests in cities, which enables efficient resource planning for future incidents. Prior work related to non-emergency incidents use simple models like gradient boosting regression, random forests and gaussian conditional random fields for solving prediction problems, that do not effectively capture the complex dependencies in the data. In contrast to the previous work, we design a deep learning based model that captures the complex underlying pattern in the data and accurately predicts future response time for non-emergency requests based on historical data. Our model is an encoder-decoder sequence-to-sequence Long-Short Term Memory (LSTM) based Recurrent Neural Network (RNN). We perform extensive experiments on the publicly available NYC 311 service requests provided by NYC Open Data. We effectively preprocess the data to deal with missing values and outliers since it makes the prediction task challenging. We use Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE) as our performance metrics. We anticipate that our LSTM based model accurately predicts the future response times with minimum RMSE and MAE values.

Additional Abstract Information

Presenter: Jillian Magyar

Institution: State University of New York- Brockport

Type: Poster

Subject: Computer Science

Status: Approved

Time and Location

Session: Poster 4
Date/Time: Tue 11:00am-12:00pm
Session Number: 3673