<![CDATA[This paper explores the application of ANN (artificial neural networks) to forecast CO2 emissions in Iraq until 2028. ANNs are able to model non-linear dynamics of time series data which eventually leads to accurate forecasts without any statistical assumption about the features of a dataset. The authors developed a simple single-input feedforward ANN model with the yearly CO2 emission data from 1991 to 2023 as the input to project the future emissions using the year. Levenberg-Marquardt algorithm was used for the network training. The model performed well on the training, validation, and testing datasets with minimal error rates and R-squared values of 1, which implied that the regression demonstrated a good fit between targets and outputs. The performance of ANNs in forecasting was evaluated. The mean squared error (MSE=0.1325) and root mean squared error (RMSE = 0.3641) values were low, highly predictive of small forecasting errors. R2 is quite high (0.946), indicating the model could explain as much as 94.6% of the variances in the actual data. The mean absolute percentage error equalled 8.01%, which signifies a good forecast with less than 10% error. The forecast of 2028 shows per capita emissions reaching 3.649 tons, which may be affected by population growth, economic development, or infrastructure changes that will be put into place. Despite renewables, efficiency, and emissions control or policies the growth curve can be replaced. This model serves as a data-driven instrument for future Iraqi CO2 emissions forecasting in order to develop climate change mitigation policies which are not time series statistical assumptions. It could also be extended to other greenhouse gases and countries, which is possible. This paper shows that ANNs can predict emissions that are accurate and reliable for decision-making which helps to reduce the country's carbon footprint and climate change.]]>
