- Select a small random subset of the original data.
- Fit a hypothesis to that sample
- Analyze the hypothesis with the rest of the data, classify which data points are inliers and outliers
- Refit the model using both the hypothesized and classified inliers
- Evaluate the refit model in terms of the residual error with respect to all of the inliers
- Repeat 4-5 as needed (This is a optimization of the model), once ready, return the model and its residual error
You do steps 1-6 on multiple samples of random data until you find a model you are happy with