Abstract:
A vast amount of diversified data from sensors, service platforms, and environmental monitoring systems has resulted from the quick development of smart cities and Internet of Things (IoT) infrastructure. IoT devices provide real-time data collection. However, the majority of structured data is processed by current analytics techniques, which restricts the identification of valuable insights required for real-time urban decision-making. This gap drives the creation of an AI-based content analytics platform to enable informed city administration by analyzing a range of smart city data streams. The thesis focuses on three key areas of the general idea of AI-based content analytics for IoT-driven smart cities: air quality monitoring and prediction, traffic congestion and detection, and citizen complaint analytics. The METR-LA traffic dataset is used to detect congested traffic using Random Forest, Decision Tree, K-Nearest Neighbours, LSTM, and CNNLSTM models. Accuracy, precision, recall, F1-score, MAE, and RMSE are assessed. Support Vector Machine, Random Forest, and Decision Tree classifiers are applied to the Chicago 311 service request dataset, and citizen complaint analytics are evaluated using accuracy, precision, recall, F1-score, and confusion matrix analysis. Machine learning and deep learning regression models for monitoring and forecasting air quality are assessed using the Beijing PM2.5 data set. We utilize machine learning and deep learning regression models for monitoring and forecasting air quality using the Beijing PM2.5 data set. The models are assessed using R, RMSE, and MAE. The models chosen depend on the task and modality. Sequence models may identify temporal patterns in traffic, tree-based models work well for classification on tabular data, and the granularity and feature representation of the data determine how well deep learning models perform for regression.
