Deep Learning for Traffic Congestion Forecasting: Unveiling Predictive Models for Enhanced Urban Mobility Planning

The project will incorporate Supervised Learning concepts using Deep Neural Networks (DNN) to forecast traffic congestion events using a public data set. A combination of Long Short-Term Memory (LSTM) and Convolutional Neural Networks (CNN) will be developed and trained using a data set from Kaggle. Finally, we will develop a comprehensive analysis of fine-tuning the hyperparameters of the DNN model. The project is based on developing and applying deep-learning models using a real-scenario data set to train a supervised learning model to forecast traffic congestion events. Artificial neural networks (ANN) as supervised learning algorithms for regression and classification problems have been extensively used in transportation management and in multiple areas of civil engineering. As presented in (Pesantez et al., 2020) for forecasting methods, ANN can effectively learn the patterns of non-linear input data to predict the correct output with data at different temporal resolutions. Furthermore, ANNs can be enhanced by the implementation of deep neural networks, specifically, Long Short-Term Memory (LSTM) and Convolutional Neural Networks (CNN) where multiple kernels define the best weight values in the learning process (Samek et al., 2021). Instead of a fully connected neural network where all inputs are connected to the nodes in the hidden layers, deep neural networks use multiple pooling layers to analyze each input data point before feeding the hidden layers (Figure 1). This research project will train LSTM and CNN as the convolution process reduces the number of parameters and improves the neural network accuracy (Gu et al., 2018). The input data comprises the traffic data from the analytics company INRIX that estimates traffic congestion cost U.S. commuters $305 billion in 2017 due to wasted fuel, lost time, and the increased cost of transporting goods through congested areas (Kaggle Data). Given the physical and financial limitations around building additional roads, cities must use new strategies and technologies to improve traffic conditions.