Outlier detection helps you spot unusual data points that might skew your findings. Use this tool when you want to identify extreme values in a dataset to decide whether to investigate, exclude, or adjust them before performing your main analysis, ensuring your results reflect the true patterns in your data.
Understanding Your Data Anomalies
Every dataset contains points that fall outside the typical range. While some of these are valid, others may result from measurement errors or unique events. Outlier detection allows you to systematically isolate these points so you can make informed decisions about their impact on your analysis.
Rather than guessing which points to ignore, this tool provides clear metrics, such as individual scores and directions, helping you distinguish between minor fluctuations and significant departures from the norm.
Choose Your Detection Strategy
Lattice offers two distinct ways to define an outlier. The IQR method uses the Tukey fence approach, which is ideal for data that does not follow a bell curve and might have a heavy tail. By looking at the spread between the 25th and 75th percentiles, it identifies points that are statistically distant from the bulk of your observations.
Alternatively, the Z-score method measures how many standard deviations a point lies from the average. This is a common approach for datasets that exhibit a symmetric, near-normal distribution.
Taking Action on Outliers
Once you have identified anomalies, you have several ways to proceed. You can choose to drop the outliers entirely, keep them as they are to maintain the full scope of your data, or use winsorization to cap extreme values at a specific threshold. These actions allow you to see how your descriptive statistics—such as the mean, standard deviation, and range—change before and after processing.
Because these actions generate a new derived dataset, you can always compare your results against the original data to ensure your modifications improve the quality of your insights rather than masking important patterns.
1 · Intent → method
An LLM picks data_outliers from a fixed catalog.
2 · Method → numbers
Deterministic Python engine runs the math. Same input → same output.
3 · Numbers → plain language
A second LLM translates the result into your domain’s vocabulary.
Will using outlier detection change my original file?
No. Lattice follows a strict non-destructive policy. The outlier detection tool creates a derived dataset for your analysis, leaving your original uploaded file untouched.
Should I use IQR or Z-score for my data?
Use the IQR method if your data is skewed or has a non-normal distribution, as it is more resilient to extreme outliers. Use the Z-score method if your data is roughly symmetrical and near-normal.
Tool input schema
Schema for data_outliers not exported yet (run pnpm export:registry).