The paper introduces a deep learning-based approach called ST-ResNet for forecasting crowd flows in each region of a city. The method aims to predict both inflow and outflow of crowds, which is crucial for traffic management and public safety. ST-ResNet is designed to handle the complex dependencies in spatial, temporal, and external factors affecting crowd flows. The model employs a residual neural network framework to capture temporal closeness, period, and trend properties of crowd traffic. Each property is modeled by a separate branch of residual convolutional units, which also models spatial properties. The outputs of these branches are dynamically aggregated based on data and external factors like weather and day of the week. Experiments on Beijing and New York City datasets show that ST-ResNet outperforms six well-known methods, demonstrating its effectiveness in predicting crowd flows.The paper introduces a deep learning-based approach called ST-ResNet for forecasting crowd flows in each region of a city. The method aims to predict both inflow and outflow of crowds, which is crucial for traffic management and public safety. ST-ResNet is designed to handle the complex dependencies in spatial, temporal, and external factors affecting crowd flows. The model employs a residual neural network framework to capture temporal closeness, period, and trend properties of crowd traffic. Each property is modeled by a separate branch of residual convolutional units, which also models spatial properties. The outputs of these branches are dynamically aggregated based on data and external factors like weather and day of the week. Experiments on Beijing and New York City datasets show that ST-ResNet outperforms six well-known methods, demonstrating its effectiveness in predicting crowd flows.