<p>The ability to predict delays has become a crucial challenge for commercial firms, since it directly influences their competitiveness in increasingly demanding markets. Therefore, in this paper, we propose a model that integrates Stochastic Petri Nets with the adaptive multi-agent systems paradigm, and this integration is further enhanced by a K-NN-based learning tool as well as a density truncation-based prediction method. Specifically, the model, named k-Nearest Neighbors Stochastic Petri Nets (K-NNSPN), introduces specialized capabilities for specifying, verifying, and analyzing complex systems driven by random events. In particular, these capabilities are demonstrated through the introduction of new token types: adaptive agents, which learn using the KNN algorithm, and prediction agents, which are responsible for performing delay prediction operations. In addition, the model is enhanced with a parameterized density truncation function, which incorporates a reduction factor for the remaining time required to complete the current task, as specified in a case from an event log. Finally, to validate the effectiveness and consistency of our model, we demonstrate its application through a real case study and simulation in MATLAB, using two datasets previously used in related work. Consequently, this comparison highlights the significant contributions of our approach in enhancing delay prediction accuracy.</p>