DeepER: A Deep Learning based Emergency Resolution Time Prediction System
| Autor: | Gissella Bejarano, Xianzhi Luo, Arti Ramesh, Adita Kulkarni, Anand Seetharam |
|---|---|
| Rok vydání: | 2020 |
| Předmět: |
050210 logistics & transportation
Mean squared error business.industry Computer science Deep learning 05 social sciences 02 engineering and technology Missing data Machine learning computer.software_genre Recurrent neural network 0502 economics and business Outlier Linear regression 0202 electrical engineering electronic engineering information engineering 020201 artificial intelligence & image processing Artificial intelligence Autoregressive integrated moving average Time series business computer |
| Zdroj: | iThings/GreenCom/CPSCom/SmartData/Cybermatics |
| Popis: | Accurately predicting resolution time for emergency incidents is crucial for public safety and smooth functioning of cities as it helps in planning resources that will be available for immediate assistance. In this paper, we present DeepER, a deep learning based emergency resolution time prediction system that predicts future resolution times based on past data. DeepER is an encoder-decoder based sequence-to-sequence model that uses Recurrent Neural Networks (RNNs) as the neural network architecture. The basic cell in DeepER is a Long Short-Term Memory (LSTM) cell. We perform experiments on the NYC Emergency Response Incidents data provided by NYC Open Data. We effectively preprocess the data to deal with uneven distribution of resolution times, outliers, and missing values. We compare the performance of the model with ARIMA and Linear Regression using two metrics— Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE). DeepER achieves an average performance improvement of 3% and 16% with respect to RMSE and 10% and 27% with respect to MAE over ARIMA and Linear Regression, respectively. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|