The fundamental reason for model retraining is that the outside world that is being predicted keeps changing and consequently the underlying data changes, causing model drift. In order to keep our predictions accurate and useful to the client, we need an efficient way to keep our models up to date. Some of our applications at IntelliSense.io such as the Heap Leach Optimization application have a number of machine learning models that constantly need to be updated and/or retrained in order for them to be useful. Building trust and rapport with our clients begins by having better transparency of model inputs and outputs.
Model drift is intended to serve as a model agnostic metric that can be applied to all of our models to give clients insight into how our models are behaving and performing over time. The tool will increase our understanding of how the input and output data to our models is changing and morphing over time.
Improved model accuracy and better performance of client processes.
Better transparency and insight to clients on model performance.
Efficient use of resources by retraining when necessary.