Predicting carbon credit prices in the carbon emission trading (CET) market is challenging due to factors such as non-stationary, non-linearity, and multi-collinearity, driven by external influences such as energy markets, economic conditions, and environmental factors. This study introduces a multi-factor prediction model, ICA-RF, which combines independent component analysis (ICA) and random forest (RF) algorithms. ICA is first applied to transform the data, comprising carbon credit prices and 22 influencing factors, into independent components (ICs). The optimal number of ICs is identified through a grid search with cross-validation, minimising MSE. These ICs are used as input for the RF algorithm to build the prediction model. Using Australian Carbon Credit Units (ACCUs) data, ICA-RF was compared with six machine learning algorithms (decision tree, Gaussian process regression, gradient boosting, k-nearest neighbours, linear regression, and Neural Networks) using five evaluation metrics. ICA-RF demonstrated superior performance, achieving an MSE of 0.291, RMSE of 0.540, MAE of 0.258, \(R^2\) of 0.998, and an execution time of 0.55 s.