Accurate classification across diverse domains is essential for information decision-making and effective resource allocation. With this, comes the Dynamic Probabilistic Mapping Model (DPMM), a flexible hierachial framework designed for multidomain predictions by analyzing feature-outcome correlations. DPMM employs a two-tiered architecture: a primary model initially categorizes each conclusion separately using one-hot encoded features mapped through probabilistic distributions. To address misclassifications and overlapping characteristics, the framework dynamically merges related classes based on performance metrics derived from confusion matrix analysis, subsequently deployed specialized subclass models for refined predictions.

This hierarchial approach enables DPMM to adapt to verying data distributions and feature interactions, enhacing classification accuracy and reliability across multiple application areas such as healthcare, finance, and cybersecurity and allowing for the creation of domain specific AI agents. For instance, in an example healthcare scenario, the primary model achieved an intiial accuracy of approximately 76% which can improve to around 87% after merging related classes. Subclass models further refine predictions, significantly boosting accuracy for specific condition groups. These results exemplofy DPMM’s capability to continuously optimize its structure based on input data dn outcome distributions. By integrating probabilistic feature mappings with dynamic class restructing, DPMM offers a robust and scalable solution for complex multi-class prediction tasks, ensuring higher precision and adaptability in diverse real-world applications.