I recently filmed a video on how to use Alteryx to build predictive models for the Institute of Actuaries Australia for a competition they’re running called Weapons of Mass Deduction. My video was shortlisted as a finalist.
Check it out, like it and vote for me if you can.
The video demonstrates how to use the predictive capabilities of Alteryx to build and fit models. I’ve used a Kaggle dataset from a competition run 8 or so years ago. The dataset tracks the success of past grants for funding. The models I’ve built aren’t great but the video does give you a rough idea of how to do this.
I like to make my workflows look as neat and organised as possible. A bit OCD like that.
I didn’t compare my results on Kaggle after putting this together, I doubt it would have performed that well. I purely used this dataset for demonstration purposes only. I find that to place high on kaggle leaderboards you need to use bespoke packages in R or Python like XGBoost or H20.ai etc. It would be difficult to produce really high performing models using the standard packages in Alteryx. That said the packages in Alteryx still allow you to produce pretty decent working models and are super easy and quick to set up and get working compared to building it in Python or R. It’s great for doing a POC and iterating fairly quickly before you invest time and effort into building the best model possible using other methods.
The example I’ve shown here is pretty basic and I didn’t delve into building an ensemble model although that would be a relatively easy way to eke out more performance.
I also don’t focus too heavily on parameter selection and tuning to build the best possible model in this video although it’s relatively easy to do in Alteryx. I keep it pretty general and high level.