Creating transparent models

One of the reasons AI has become increasingly popular over the past decade is the development of more complicated models, such as deep learning models. Deep learning models are especially successful on unstructured data as they can derive what features are needed to generate the prediction.

The benefit of deep learning models is that they are often more accurate, while the disadvantage is that they tend not to be very transparent: it is unclear how the model generates a prediction or makes a decision.

Using algorithms that are transparent by design

Transparency is becoming an increasingly important concern when it comes to training machine learning models. Even though more complicated algorithms can be used to train more accurate models, sometimes, a data scientist may opt for a simpler algorithm that is more transparent. The following diagram shows how simpler algorithms, such as linear models and decision trees, have better transparency but may...