Adopting a data-first approach for explainability
In Chapter 3, Data-Centric Approaches, we discussed the importance and various techniques of Data-Centric XAI. Now, in this section, we will elaborate on how adopting a data-first approach for explainability helps in gaining users' trust in industrial use cases.
Data-centric AI is based on the fundamental idea that the quality of the ML model is as good as the quality of the underlying dataset used for training the model. For industrial use cases, dealing with poor-quality datasets is a major challenge for most data scientists. Unfortunately, data quality is often ignored as data scientists and ML experts are expected to cast their magic of ML to build models that are close to 100% accurate. Consequently, ML experts simply try to follow model-centric approaches such as tuning hyperparameters or using complex algorithms to boost model performance. Even if the model performance increases slightly, with the increase in complexity...