Improving the accuracy of short-time traffic prediction in intelligent transport system based on machine learning algorithms

Abstract

Intelligent traffic management systems, urban planning, and the reduction of traffic congestion all depend on traffic flow prediction. In this research, a method based on machine learning based on neural network combination and feature selection based on genetic algorithm is presented for predicting short-time traffic flow. The genetic algorithm-based approach seeks to find a model's optimal parameters globally. Inner-city traffic constantly changes and can be unpredictable. This is because traffic patterns repeat over time (have periodic characteristics) but also swing wildly from moment to moment (high fluctuations). As a result, it's very hard to guess what traffic will be like in the future. Thin operators have been used duo to it good performance for short-time traffic prediction in neural networks system. In Isfahan gathered traffic data to see how well a new model, called LSTM, predicts traffic flow. We compared LSTM's performance against other established methods like wavelet neural networks (WNN) and multilayer perceptron (MLP). the proposed neural network prediction model and genetic algorithm results have %97 accuracy, %97 correlation coefficient, 14.67 less average absolute error, higher signal-to-noise ratio, 0.97 entropy value, and 3.95 standard deviation. Compared to other methods, it has shown its superiority, such as ordinary neural networks. This model excels at finding the best solution quickly and accurately, even with noisy or complex data.