ML models come in many flavours. Some models are naturally easy to explain. Rules-based models or simple statistical ones can be easily inspected and intuitively understood. The typical machine learning approaches are usually harder to understand. At the extreme, deep learning models are very complex and need specialised approaches to understand their inner workings.
It is important to know if explainability to end users is a requirement up front, because this will influence the model you decide to go with. In some use cases, there is a regulatory need and explainability is an essential requirement e.g. for credit risking it is essential to be able to explain why an applicant has been denied a loan. In other cases the model will simply not be accepted by end users if they cannot understand how a decision has been reached.
Explainability goes hand in hand with simplicity, and a simple model may well perform worse than a complex one. SIt is common to find that an explainable model performs less well in terms of accuracy. This is fine! Accuracy alone is not always the only measure of a good model.
In our experience, engaging the end user and explaining how the model is making decisions often leads to a better model overall. The conversations you have with end users who understand their domain and data often result in the identification of additional features that, when added, improve model performance. In any event, explainability is often a useful part of developing the model and can help to identify model bias, reveal unbalanced data, and to ensure the model is working in the intended way.
If you find you have a complex model and need an explanatory solution, these tools can help:
SHAP - good for use during model development
The What If Tool - good for counterfactual analysis
Google’s AI Explanations - good for using on deployed models (tensorflow models, tabular, text, image data, AutoML)
In one of the engagements that I was involved in, we had to take an ML initiative from an existing partner. During discovery we found that the stakeholders in the business units did not understand the machine model that had been built and their questions related to the output (predictions) were answered with deep technical jargon that they could not comprehend. This resulted in the business units at best using the output grudgingly without any trust or at worst completely ignoring the output.
One of the first things we did when we took over the operations of the system was to translate the model outputs into outcomes and visuals that explained what the model was predicting in business terms. This was done during the initial iterations of building the model.
Three significant changes happened in how the data team was able to collaborate with the business units:
The stakeholders understood what the model was trying to do. They were able to superimpose the output of the models on their own deep understanding of the business. They either concurred with the model outputs or challenged them. The questions that they raised helped the data team to look for errors in their data sources/assumptions or explore additional data/features, thereby improving the output.
The business units also understood better the need for high quality data that affect the model outputs. They took steps to fix processes that either were incomplete in data collection or were ambitious resulting in confused data collection.
As the stakeholders were involved very early in the model building process, they considered themselves to be co-creators of the model rather than just consumers. This resulted in enthusiastic adoption of outputs including acceleration of any process changes needed to leverage the work.
We were able to deploy five models in production over a period of six months that were being used to generate business outcomes compared to one model that went live in the earlier attempt, after 18 months. Oshan Modi Data scientist
Equal Experts, India