DETECTING SOCIAL EVENTS IN SPATIO-TEMPORAL DATA USING CONV-LSTM AND RESNET-50

Mohammad Hani, Pak Vadim G.

The article focuses on event detection by analyzing data related to space and time. It examines the use of neural networks to process numerical data from mobile communications companies, with the goal of discovering social activities. The properties of communication data, particularly their connection to time and location, enable the prediction of potential social events in specific places and times. Recent advancements in deep learning have significantly improved predictive capabilities. Many studies have used deep models, such as LSTM neural networks, to detect anomalies, but these often lack consideration of spatial features or are based on convolutional neural networks (CNNs) alone. However, no previous research has applied a combination of ConvLSTM-based neural networks and ResNet50 to this type of data.

Keywords:spatial-temporal, ConvLSTM, ResNet50, Mahalanobis distance, event detection.