IQR Outlier Comparison

Inspiration

With so many top AI models available, it’s hard to know which gives the most reliable results. We wanted a way to quickly spot and compare which models produce unusual or extreme outputs, enabling better, data-driven choices.

What it does

IQR Outlier Comparison analyzes and compares output from multiple leading AI models, using the statistical Interquartile Range (IQR) method to find outliers. It visualizes which models are the most consistent, and which produce the most unusual results, making model selection more transparent.

How we built it

We collected outputs from major AI models and applied the IQR method to identify outliers for each. Using Hex, we built a dashboard to automate the analysis, aggregate the results, and display clear summaries and visualizations.

Challenges we ran into

Gathering a large and diverse sample of model outputs. Ensuring fair comparison across different models and output types. Tuning the IQR method to best capture outlier behavior in each context.

Accomplishments that we're proud of

Built a single, interactive dashboard to compare 15 major models at once. Automated the outlier computation and visualization. Delivered actionable insights to help users pick the most reliable AI models for their needs.

What we learned

Outlier frequency can vary widely across models and tasks. Even top-tier models sometimes produce surprising results. The IQR method offers a clear, objective way to compare model stability.

What's next for IQR Outlier Comparison

Add support for more models and larger datasets. Enable custom threshold selection for advanced users. Integrate with real-time model outputs for live monitoring and alerts.

Built With

• hex