Machine Learning models have demonstrated immense success in making accurate predictions from unobserved data. However, many of the successfully deployed models that have high predictive accuracy are usually black box models. Judging performance primarily based on their predictive accuracy potentially lead to technical debt and other fundamental issues, especially in areas of application such as security, nondiscrimination, customer retention, etc.
The objective of this talk is to enlighten the audience about some of the pitfalls in using widely popular ML algorithms and offers some recommendations to address these concerns.