
Accelerator Synopsis
The ML models are usually BlackBox. Explainable AI (EAI) allows users to probe their trained models, understand and interpret predictions from machine learning models. Further, the ML practitioners can use this platform to generate feature attributions for model predictions and visually inspect model behavior.
EAI provides additional capabilities for end-to-end exploratory data analysis of training and scoring data sets to combine data with insights and further verify the explanation with the help of analysis. It also proves to be a convenient plug-and-play in feeding model objects as inputs while reducing downtime during setup.
Value Addition

Model Probing
EAI allows users to test model performance between defined target classes and within the local scope. It analyzes various data features and uses counterfactuals in several simulation scenarios to visualize model behavior.

Statistical Explanation
Provides visually fascinating EDA methods to understand data trends & distributions better with data description, sample view, & correlation analysis. Users can perform univariate/outlier analysis and bivariate analysis on the data
Features

File Upload
Access the project log, upload new model objects and create new projects for the explanations. Users can upload both training and scoring data sets.

Explainability
Global explainability indicates the percentage contribution of a target class to the prediction value. Local explainability provides the percentage contribution of the feature to the prediction value.

Simulation
Helps understand how the predicted value changes by making edits to actual value from any point on the dataset by accessing a scatter plot data distribution.

Counterfactual
This view is only applicable to classifiers. It shows the counterfactual data point closest to the actual point selected but from a different class.

Data Description
Data sample view contains the top 10 rows, the correlation matrix indicates the degree of dependency between variables, and data feature information provides data statistics.

Data Distribution
Provides bivariate analysis of continuous variables and bivariate analysis of continuous vs. categorical variables, class balance, variable distribution.

Outlier Analysis
Provides univariate analysis of continuous variables in the data set. The statistics of the Boxplot chart include lower limit, Q1, median, Q3, and upper limit.
Sparking AI Innovation at the Intersection of Business Analytics, Data Science, and Engineering.



FAQs
Delivery team, data scientists, and business analysts. Anyone who needs to understand the performance of their AI/ML models.
Currently, EAI supports tabular data sets. In the future edition, text and image records will be introduced.
Classifier and Regression models: Random Forest, XG Boost, DNN, Logistic Regression, Light Gradient Boost, ANN (Artificial Neural Networks).
XAI Tabular Library for Neural Network Models and Interpret Community Library for the other models.
- Input Data Format – Pickle file (.pkl, .sav, .jsm formats).
- EDA – bivariate, univariate analysis, outlier analysis, correlation analysis, etc.
- Explainability – Global and local explanation
- EDA Simulator – New scenario, counterfactual value

