Compare scores using PCA, NNet, and linear regression models

[rayJ]

First: the question - how to read output of a NNet model and compare it with other models such as LR model.

I am doing this weekly challenge #18. The goal is to find out projected wins, losses, and games back for the next year.

I've tried to solve it using three models: 1)PCA using top 10 variables, and then use NNet mode with 3 PCsl; 2)PCA using top 10 variables and then Linear regression using 3 PCs; and 3)the official solution: linear regression using top 10 variables directly

The three solutions give three different outputs and I cannot decide which solution is better. I can certainly compare the MSE and RMSE from LR. However, those information is not available using NNet model.

I've included the workflow and the input data file. The weekly challenge's link is down below:

https://community.alteryx.com/t5/Weekly-Challenge/Challenge-18-Predicting-Baseball-Wins/td-p/36747/p...

Thanks in advance

RayJ

[BrandonB · Accepted solution]

It looks like another individual had a similar idea. You can take a look at his solution to check your work: https://community.alteryx.com/t5/Weekly-Challenge/Challenge-18-Predicting-Baseball-Wins/m-p/86248/hi...

You probably want to take a look at the Model Comparison macro that you can download from the Predictive District. This is an Alteryx created tool but it is not installed with Designer by default. This tool will allow you to compare the NNet as desired: https://gallery.alteryx.com/#!app/Model-Comparison/5d92741a826fd30b84537a27

Here is a list of many other predictive tools that are also available for download: https://community.alteryx.com/t5/Alteryx-Designer-Knowledge-Base/Predictive-Tools-Available-Outside-...