Metaheuristic-optimized neural networks for travel time index prediction: A comparative study of wild horse and coot optimization algorithms

• This study pioneers the use of Wild Horse Optimization (WHO) and Coot Optimization Algorithm (COA) to optimize Multilayer Perceptron (MLP) neural networks for predicting the Travel Time Index (TTI). • The WHO-optimized MLP outperforms the COA-optimized model, achieving lower Mean Squared Error (MSE: 0.0026 vs. 0.0028). • WHO demonstrates faster convergence with a computational time of 147 seconds compared to COA’s 160 seconds. • The MLP, configured with four input variables (traffic volume, weather, crash rate, section length), one hidden layer with 12 neurons, and one output neuron, effectively captures nonlinear traffic dynamics. Accurate travel time prediction is essential for effective Intelligent Transportation Systems (ITSs). Unlike traditional gradient-based methods, which may get trapped in local optima, metaheuristic algorithms efficiently navigate complex optimization landscapes without needing derivative computations. This study compares two metaheuristic algorithms—Wild Horse Optimization (WHO) and Coot Optimization Algorithm (COA)—for optimizing the weights and biases of a Multilayer Perceptron (MLP) neural network to predict the Travel Time Index (TTI). The MLP uses four input variables (traffic volume, weather, crash rate, and section length), one hidden layer with 12 neurons, and one output neuron. The model was trained and tested on a dataset of 472 observations from Virginia’s transportation network (2014–2017), with input variables showing mean values of 1.19 (traffic volume), 6.92 mm (precipitation), 0.17 (crash rate), and 2.76 miles (section length), and standard deviations of 0.20, 5.21, 0.12, and 1.64, respectively. Performance was assessed using Mean Squared Error (MSE), Root Mean Squared Error (RMSE), and Correlation Coefficient (R). The WHO-optimized model outperformed COA, achieving an MSE of 0.0026, RMSE of 0.0511, and R of 0.7516, compared to COA’s MSE of 0.0028, RMSE of 0.0533, and R of 0.7262. WHO also showed better computational efficiency (147 seconds vs. 160 seconds for COA). These results highlight the WHO-MLP model’s superior accuracy and speed, making it ideal for real-time navigation systems.