Explainable AI

When the dashboard is created, the first section is the XAI tab which provides the overall explanations of the model and performance metrics.

📌What Is an Explanation?

An explanation is the answer to a why-question (Miller 2017).

• Why did not the treatment work on the patient?

• Why was my loan rejected?

This section is divided into 3 sections:

Summary

Here, you'll find an initial snapshot of the dataset and a high-level overview of the global behaviour. Under it is Feature Influence, indicating the correlation each feature has, differentiating positive from negative impacts and then finally at the bottom, the Global Feature Importance, a chart ranking various features by their importance.

Feature Attributions

In the Feature Attributions tab, you can access in-depth analyses of how individual features influence model predictions.

These plots quantify the contribution of each feature to the prediction, allowing for precise interpretation of feature influence. Additionally, you'll find density distribution plots that visualize the prevalence of each feature state across the dataset.

📌Pinning and Sharing Insights

You can pin plots and add them to the your sheets in the Analyze tab. Simply hover over the chart and click the '+' icon.

Why should I analyze feature attributions?

1. Identifying Important Features: By analyzing feature attributions, you can identify which features are most influential in driving the model's predictions. For example, if a feature like CustomerLoyaltyScore strongly impacts customer retention, businesses can focus on enhancing loyalty programs and offering personalized discounts to improve retention rates and boost sales.

2. Detecting Data Biases: Feature attributions can reveal biases or inconsistencies in the dataset. If certain features consistently have a disproportionate impact on predictions, it may indicate biases in the data or the model itself. Addressing these biases can lead to fairer and more reliable predictions.

3. Improving Model Performance: Insights from feature attributions can guide model refinement efforts. By focusing on features that have the most significant impact on predictions, data scientists can prioritize feature engineering, tuning, or model selection to improve overall performance.

Overview

The next tab, Overview is where you can see the metrics and quality issues, such as high cardinality features, missing features and more.

We use the following metrics at View AI,

• For Binary Classification Models:

  • Accuracy

  • True Positive Rate/Recall

  • False Positive Rate

  • Precision

  • F1 Score

  • AUC

  • AUROC

  • Binary Cross Entropy

  • Geometric Mean

  • Calibrated Threshold

  • Data Count

  • Expected Calibration Error

• For Multi-Class Classification Models:

  • Accuracy

  • Log loss

• For Regression Models:

  • Coefficient of determination (R-squared)

  • Mean Squared Error (MSE)

  • Mean Absolute Error (MAE)

  • Mean Absolute Percentage Error (MAPE)

  • Weighted Mean Absolute Percentage Error (WMAPE)

↪ Questions? Chat with an AI or talk to a product expert.