Data-Driven: This approach involves using algorithms that can extract the desired information directly from the sensor data. It involves machine learning methods including linear regression, neural networks, random forest, and hidden Markov models. This approach can be used when there is no background information available to define a mathematical equation.
Model-Driven: This approach uses the available physical background information or system behavior to define the mathematical equation by combining the sensor data to predict an output using algorithms such as Kalman filter (KF), particle filter (PF), and others.
Additionally, Roofline is a technique used to identify performance bottlenecks in models on targeted computing platforms. This method captures the compute-memory ratio and helps to identify if computation is memory- or compute-bound.
