Improving LSTM Framework Based on Hybrid Meta-heuristic Optimization Techniques for Traffic Flow Forecasting

Short-term traffic flow forecasts are important for travel safety, congestion avoidance, and traffic management and are an integral part of intelligent transport systems (ITS). Long short-term memory (LSTM) has become a promising technique for forecasting traffic flow. Unfortunately, the LSTM model does not achieve adequate forecast accuracy due to both noisy traffic data and poor selection of hyperparameter optimization values. To improve the convergence and accuracy of short-term traffic stream forecasts, a composite model consisting of LSTM neural network trained and tuned using sinusoidal firefly feature selection (SFFS) and dragonfly (DF) algorithms has been developed. This model efficiently determines the optimal set of LSTM features and hyperparameters for domain prediction of traffic flow while minimizing training errors. The SFFSDF-LSTM network model is used to find the ideal weights, biases, and features of LSTMs for better convergence and lower error rates. From the perspective of error analysis and predictive analytics, the predictive accuracy of the combined model is evaluated and found to be superior to traditional non-parametric models. Hosted fileSFFSDF-LSTM₂5. 08. 2024. docxavailable at https: //authorea. com/users/826111/articles/1221432improving-lstm-framework-based-on-hybrid-meta-heuristic-optimization-techniques-fortraffic-flow-forecasting