<p>In artificial intelligence (AI), linear models are widely used to establish relationships between target and input variables. The traditional least squares method, rooted in classical statistics, assumes that both the training data and test data are precise and certain. However, this assumption may not hold in situations involving uncertainty. This paper presents an extension of the linear model under neutrosophic statistics, allowing AI models to be trained and tested while accounting for uncertainty through the incorporation of the degree of indeterminacy. We introduce the neutrosophic linear model and its associated matrices. The proposed model is utilized to train a model using neutrosophic data, with the resulting parameters subsequently applied to evaluate the model’s validity. We conduct comprehensive simulation studies to examine how varying degrees of indeterminacy affect the linear model’s matrices and the predicted values from both training and testing phases. Furthermore, a practical application using neutrosophic data from the education sector is provided. Our findings emphasize the importance for AI decision-makers to consider the degree of indeterminacy during model training and testing to effectively address uncertainty.</p>