Sensor Data Explainability for Facility State Monitoring

The patent introduces explanation algorithms (LIME, SHAP) as intermediary components between the learned model and the user interface. These intermediaries translate the complex internal workings of the machine learning model into comprehensible explanations about which sensor data points most strongly influence facility state predictions, thereby resolving the contradiction by maintaining high monitoring accuracy while reducing the perceived system complexity through interpretability

The system implements feedback loops where the explanation results are continuously provided back to users, allowing them to understand and verify the model's reasoning. This feedback mechanism enables users to trust and effectively use the complex learned model without being overwhelmed by its complexity, as they can see which specific sensor readings are driving the facility state assessments

The patent applies partial action by selecting and explaining only the most influential sensor data points rather than analyzing all input data. The explanation algorithms identify and highlight the top contributing factors to facility state predictions, providing sufficient interpretability without the need to process and explain every single data point, thus balancing interpretability with computational efficiency

The system dynamically adjusts parameters such as the number of explanations generated, the granularity of detail provided, and the threshold for significance when applying LIME and SHAP algorithms. By changing these parameters based on computational resource availability and user needs, the system can optimize the balance between providing comprehensive interpretability and maintaining acceptable processing times