Dataiku and Explainable AI: understanding Artificial Intelligence decisions

Explainable AI Dataiku


Artificial Intelligence (AI) is rapidly becoming an integral part of our lives, influencing various sectors including healthcare, finance, commerce, transportation, and more. However, with the increasing impact of AI, there arises the need to make this technology more understandable and transparent in order to build trust in the technology itself. The exponential growth in the performance of AI models is accompanied by an increasing need to interpret the results produced and to explain the decisions made by the algorithms. For these reasons, Explainable AI (XAI) is becoming increasingly crucial. In this context, tools like Dataiku emerge as fundamental resources, enabling transparent and responsible management of Artificial Intelligence.

What is Explainable AI?

Models like linear regression or decision trees inherently possess great explainability, and the results produced are easily interpretable. On the other hand, more complex and high-performing models like neural networks are much more difficult to understand and interpret. For this reason, such models are often referred to as “black box.” In choosing between these algorithms, it is necessary to address the so-called trade-off between accuracy and interpretability. Explainable AI (XAI) fits into this context, encompassing algorithms, methods, and processes that help explain how AI models work in a clear manner and provide tools to interpret even the results of complex models. Explainable artificial intelligence is one of the necessary elements to implement responsible AI solutions, which are systems based on transparency and user trust, taking into account ethical implications. For further insights on these topics, we suggest reading our articles on Explainable AI and Responsible AI.

Explainable AI

Dataiku Tools

Dataiku, a cutting-edge platform in the field of data science and machine learning, offers a range of features that support Explainable AI. Through the integration of various techniques and tools, Dataiku provides a robust approach to interpreting and explaining the complex predictions of models, providing users with a clear analysis of influential variables and decision paths.

XAI tools are integrated into machine learning solutions and are accessible to both technical and non-technical users. Additionally, the platform supports users in creating documentation by automatically generating files with information about the models and their construction. Below are some of the tools provided by Dataiku to create Explainable AI solutions.

  • Feature Importance: every machine learning model provides a graph measuring the impact of each variable on the model’s predictions. Moreover, it is possible to view this indication for individual points in the test dataset and as a function of a single variable.
  • Partial dependence plot: this type of graph allows you to understand how the probability of belonging to an outcome class changes as a certain variable varies. In a churn analysis, partial dependence would help answer questions like “How does the probability of customer churn change with age?”
  • Subpopulation analysis: this tool provides insight into the model’s performance for different subsets of data. For example, we could see if the model achieved similar predictive accuracy between males and females. This type of analysis reduces the risk of deploying models that are biased for a certain subpopulation.
  • Individual prediction explanations: through Shapley values or the Individual Conditional Expectation (ICE) method, Dataiku allows you to understand which variables have the greatest impact on the prediction of outcomes for individual individuals.
  • What-if analysis and simulation: by setting input values, users can simulate one or more new observations and evaluate the model’s predictions and the variables that have the greatest impact on the new data.

Why Implement Explainable AI?

In conclusion, Explainable AI is essential to ensure transparency and accountability in the use of Artificial Intelligence. Dataiku is at the forefront of this effort, offering tools and methodologies to integrate XAI into business decision-making processes. Through the integration of various Explainable AI techniques, Dataiku provides a robust approach to interpreting and explaining the complex predictions of models, providing users with a clear analysis of influential variables and decision paths. Blue BI is a Dataiku partner, and our team of data scientists can help your company adopt interpretable and transparent AI solutions. Contact us to learn more!

We realize Business Intelligence & Advanced Analytics solutions to transform simple data into information of freat strategic value.


Table of Contents