Machine Learning and statistic predictive modeling for road traffic flow

Abstract

Traffic forecasting is a research topic debated by several researchers affiliated to a range of disciplines. It is becoming increasingly important given the growth of motorized vehicles on the one hand, and the scarcity of lands for new transportation infrastructure on the other. Indeed, in the context of smart cities and with the uninterrupted increase of the number of vehicles, road congestion is taking up an important place in research. In this context, the ability to provide highly accurate traffic forecasts is of fundamental importance to manage traffic, especially in the context of smart cities. This work is in line with this perspective and aims to solve this problem. The proposed methodology plans to forecast day-by-day traffic stream using three different models: the Multilayer Perceptron of Artificial Neural Networks (ANN), the Seasonal Autoregressive Integrated Moving Average (SARIMA) and the Support Machine Regression (SMOreg). Using those three models, the forecast is realized based on a history of real traffic data recorded on a road section over 42 months. Besides, a recognized traffic manager in Morocco provides this dataset; the performance is then tested based on predefined criteria. From the experiment results, it is clear that the proposed ANN model achieves highest prediction accuracy with the lowest absolute relative error of 0.57%.

Automated machine learning: the new data science challenge

“…Based on the amount of data and type of supervision gotten during the training process, ML systems can be devided into four main categories:i) supervised learning: in this type of learning algorithm, the problem can be either classification (logistic regression, k-nearest neighbour (KNN), support vector machine (SVM), naïve Bayes) or regression (decision tree (DT), linear regression, random forest (RF), support vector regression (SVR) [12]). It is used with labeled data whre the mapping patterns from the training set of as given model is learned from input to output.…”

“…Considering that, this study tackles the use of various methods (parametric and non-parametric [25]) to forecast traffic based on real Moroccan dataset of a toll station with high traffic volume. The approach consists on comparing, according to predefined criteria, the predictive performances of three different methods, namely: i) neural networks structure MLP [4] as a non parametric model, ii) mathematical modeling method seasonal ARIMA (SARIMA) [26] as a parametric model, and iii) SMOreg inspired from the SVM algorithm [12] as a non parametric model.…”

Towards a new intelligent traffic system based on deep learning and data integration

“…According to the study conducted by Duchenaux et al [2], increasing road capacity does not systematically lead to optimal congestion reduction, and may even causes serious traffic conditions. To achieve better traffic flow prediction performance, many prediction methods have been proposed covering a wide spectrum, such as parametric methods, non-parametric methods and hybrid methods [3]- [8].…”

Experimental Analysis of Time Series Models for Traffic Flow Prediction

“…A well-liked time series analysis method that has been applied to predict traffic is the auto-regressive integrated moving average to forecast web traffic [12] and road traffic. [13,14] The processing of complicated data is still a limitation for the single model to process. Therefore, combination models are more preferred over single models to enable more precise traffic flow forecasts.…”