<p>Predicting subsequent activities represents a critical component of business process monitoring, as it significantly enhances managerial decision-making processes, including in terms of resource allocation, compliance verification, real-time alerting, and the development of recommendations for process improvements. Previous researchers have used basic embedding techniques to incorporate various log attributes and activity information as inputs for predictive models, thereby improving the predictive accuracy of such models. However, these approaches have often overlooked the importance of distinct attributes and behaviors, as well as the interactions between data attributes and behaviors in this context. This paper presents a novel methodology known as the multiattribute collaborative filtering (MACF) model, which employs neural collaborative filtering to predict subsequent activities. Initially, the model uses self-attention networks and graph neural networks to determine the importance scores of data attributes and behavioral activities, respectively. The model subsequently integrates neural collaborative filtering with a multilayer perceptron to capture both linear and nonlinear interactions between these data attributes and behaviors. The pretrained log vectors are then fed into the neural network to predict the subsequent activity. The MACF method was implemented with the assistance of Python and evaluated by reference to six real-world event logs. A comparative analysis on the basis of various benchmark methods reveals that the MACF model exhibits superior predictive performance.</p>