Predicting New York City Taxi demand: spatio-temporal Time Series Forecasting | Fabian Hertwig

Predict taxi demand in New York City with spatial and temporal accuracy using convolutional neural networks (CNNs) and explore the challenges and limitations of real-world data application.

Key takeaways

The presentation is about predicting New York City taxi demand using spatio-temporal time series forecasting.
The data comes from the New York City taxi commission and includes information such as trip start and end locations, time, and fare.
The process model involves six phases: business understanding, data understanding, data preparation, modeling, evaluation, and deployment.
The data is cleaned and pre-processed, and then regrouped into hexagons to account for both spatial and temporal context.
The model used is a convolutional neural network (CNN) with a temporal convolutional layer, which learns patterns in the time series data.
The model is evaluated against several baselines, including a naive approach that predicts the same demand as the previous hour.
The results show that the CNN model outperforms other methods, including LSTM and ARIMA.
The model is able to capture seasonal patterns and trends in the data, and is able to make accurate predictions for future demand.
The presentation also touches on the challenges of working with taxi data, including dealing with missing values and anomalies.
The authors also discuss the limitations of the model, including the need for retraining and the use of a limited dataset.
Future work involves exploring other models and methods for spatio-temporal time series forecasting, as well as integrating the model with other data sources.

Predicting New York City Taxi demand: spatio-temporal Time Series Forecasting | Fabian Hertwig

More talks