Non - deterministic
Different behavior for the same inputs – the behavior can change with change in test / training data
Data adequacy and
quality of data
Large amount of data to train the models, and data quality in terms of bias, skewness etc.
Domain knowledge
Domain knowledge is necessary to understand data, feature engineering, results interpretability and applicability etc.
Bias
Understanding of data for strata, their representation, and sampling – human bias, noised data etc.
Data Interpretability
Data interpretability – it being a black-box, when it comes to it working or explainability of data