Enhancing CO2 emissions prediction for electric vehicles using Greylag Goose Optimization and machine learning

Abstract

Electric vehicle (EV) [Formula: see text] emissions should be predicted and mitigated, which requires lowering EV emissions in line with global sustainability goals. Such accurate forecasting supports policymakers and other industry stakeholders make marketing decisions to reduce environmental impacts and optimize resource utilization. In this research, a novel Greylag Goose Optimization (GGO) algorithm is integrated with a Multi-Layer Perceptron (MLP) model to improve [Formula: see text] emissions prediction. Finally, the study does a comparative analysis with some established optimization algorithms in hyperparameter tuning regarding an improved accuracy model. In addition, statistical analyses such as ANOVA, sensitivity analysis, and T-test were used to substantiate performance differentiation between models. For the optimal model, the GGO-optimized MLP significantly outperformed baseline models and other optimization techniques, having minimum error metrics such as correlation coefficient and RMSE and an MSE of [Formula: see text]. As a result, the emissions forecast is very reliable. The proposed approach provides actionable insights for environmental policies, EV adoption strategies, and infrastructure planning. The model enables stakeholders to achieve climate objectives, optimize EV charging systems and foster the creation of sustainable transportation systems, as said accurate emissions estimates are enabled.

Keywords: [Formula: see text] emissions prediction; Electric vehicles (EVs); Greylag Goose Optimization (GGO); Multi-Layer Perceptron (MLP); Sustainable transportation.

Fig. 1

Correlation matrix for key variables.

Fig. 2

emissions over time (2000–2022).

Fig. 3

Fuel consumption by vehicle manufacturer.

Fig. 4

Distribution of key vehicle attributes.

Fig. 5

Framework of the paper.

Algorithm 1

Greylag Goose Optimization (GGO) Algorithm

Fig. 6

Parallel coordinates plot comparing baseline regression models.

Fig. 7

Radar plot of performance metrics for baseline models.

Fig. 8

Heatmap of performance metrics for baseline models.

Fig. 9

KDE plots for the distribution of metrics across optimized MLP models. The density functions highlight variations in the metrics such as MSE, RMSE, and correlation coefficients.

Fig. 10

Bar plot comparing MSE, RMSE, and MAE for all optimized MLP models. Lower values for these metrics indicate better performance.

Fig. 11

Violin plots showing the distribution of metrics such as MSE, RMSE, MAE, , , RRMSE, NSE, and WI across all models. Each violin includes a swarm plot overlay for individual data points.

Fig. 12

Mixed plots combining boxplots and violin plots for metrics distribution. This visualization highlights the metrics’ spread and central tendency across models.

Fig. 13

Heatmap showing the comparison of metrics across models. Metrics include MSE, RMSE, MAE, , , RRMSE, NSE, and WI. Darker shades represent better performance.

Fig. 14

Summary plot with scatter points for key metrics and a heatmap showing the relative performance of optimized models. The heatmap emphasizes the superior performance of GGO-MLP and related algorithms.

Fig. 15

Histogram showing the distribution of performance metrics (e.g., MSE, RMSE, and MAE) across all trials for optimized models. The chart highlights the concentration of lower error values for the best-performing models.

Fig. 16

Scatter plot of performance metrics across optimized models. The plot highlights the relative distribution and outliers for GGO-MLP, SFS-MLP, and WOA-MLP models.

Fig. 17

Convergence time analysis for different parameter values of GGO, illustrating the computational efficiency under different conditions.

Fig. 18

Histogram of optimization results for different parameter settings, illustrating the distribution of fitness values.

Fig. 19

Histogram representation of fitness results across varying parameter configurations, highlighting performance trends.

Fig. 20

Performance variation analysis for different GGO parameter settings, demonstrating their effect on optimization accuracy.

Fig. 21

Comparison of optimization results under different parameter settings, highlighting fitness function trends.