0
$\begingroup$

I'm doing an analysis on some data involved with the NBA right now, and I'm nearing the end of the project but I need to decide if I want to include an official prediction for future data or not.

From what I've seen online, the steps to a Machine Learning project is Gathering Data, Data Prep, Observe Data, Train Model, Evaluate Model, then Predict (on unknown data).

The first 5 steps seem to all the same as a data analysis but a "data analysis project" would focus a lot more on the observe data part and derive conclusions from that. What I did was follow all the steps, derive conclusions from the data and graphs I've observed, I ran some models on the data as well to prove some correlation between the data, and to prove that a trained model can predict data.

However, at the end I don't think a prediction of the future is necessary for my project. If I don't include this "predicting" part of my analysis, can I not technically call that a "Machine Learning" project? Because all I did was look at and analyze data and graphs and derive a conclusion from that rather than concluding with a "with our trained model we can predict these trends"?

This is a big vague so let me know if I can clarify anything, thanks!

$\endgroup$
7
  • $\begingroup$ What is your plan to evaluate the model if you’re not making predictions? $\endgroup$
    – Dave
    May 7, 2020 at 3:38
  • $\begingroup$ I want to mainly observe the trends of the data with graphs and charts, and then use the algorithms like Random Forest to see if there's a model that can accurately predict outputs based on my input values. If the accuracy is pretty high then that means that the data is "valid" and the features are relevant? @Dave I guess I'm not really interested in guessing what will happen in the future and present those predictions, but I'd like to know if I COULD if I wanted to, and show that $\endgroup$
    – chung
    May 7, 2020 at 3:51
  • $\begingroup$ And how do you plan to do that? $\endgroup$
    – Dave
    May 7, 2020 at 10:20
  • $\begingroup$ @Dave Just train a model as usual with a training set and a testing set, and seeing if there's an algorithm that has a high accuracy with the stratified k-fold method since my data is classification type. Then I just record that and finish, rather than ending off with a future prediction $\endgroup$
    – chung
    May 7, 2020 at 15:58
  • $\begingroup$ How is your out-of-sample testing not different from a future prediction? (My point is that, unless I completely misunderstand what you're doing, YES it's machine learning! Anyone who applies the process you've described to the 70,000 MNIST images is doing machine learning, after all.) $\endgroup$
    – Dave
    May 7, 2020 at 16:07

1 Answer 1

0
$\begingroup$

Could you talk a bit more about your model training and evaluation?

If you are viewing correlations, it sounds like exploratory data analysis to me.

Typically, I believe a machine learning model will be trying to predict a data point/variable. This could be in the form of classification or regression (i.e. did you do one of these?)

Just curious, why is it important to label as a machine learning project? If a class project, maybe you would be interested in your model predicting points, points per game, or win/loss ratio. Or maybe based on your model, you would want to be clustering players into different segments.

Good luck!

$\endgroup$
3
  • $\begingroup$ My project is similar to taking data from previous drafts and seeing who will be good. I guess the point of my project is finding out if this specific data I collected is useful/relevant at all for using to predict the future. And yes I did use classification algorithms to train my model. The only thing is I don't really care about predicting "this player from 2020 will be good", I just want to prove that my data analysis is useful because my trained model could accurately predict which players in the past will be good using my data. So the train set & test set is all I need, not predict sets $\endgroup$
    – chung
    May 7, 2020 at 3:56
  • $\begingroup$ That definitely sounds like a machine learning project to me. By validating on a test set, you are making predictions, just not on future data. $\endgroup$ May 7, 2020 at 7:40
  • $\begingroup$ Interesting take, thanks! $\endgroup$
    – chung
    May 7, 2020 at 18:53

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.