Towards an Ontological Framework for Integrating Domain Expert Knowledge with Random Forest Classification

This paper proposes an ontological framework that combines semantic-based methodologies and data-driven random forests (RF) to enable the integration of domain expert knowledge with machine-learning models. To achieve this, the RF classification process is firstly deconstructed and converted into semantic-based rules, which are combined with external rules constructed from the knowledge of domain experts. The combined rule set is applied to an ontological reasoner for inference, producing two classifications: (1) from simulating the selected RF voting strategy, (2) from the knowledge-driven rules, where the latter is prioritised. A case study in the steel manufacturing domain is presented that uses the proposed framework for real-world predictive maintenance purposes. Results are validated and compared to typical machine-learning approaches.