Exercise: Predicting Diabetes Onset

[lx0n2acl]

When I run my linear classifier for 100 epochs I get about 76% accuracy. Increasing the epochs doesn't change it by much. Is there anything I can do to the feature set to increase my model's accuracy? I already ran the scaler... I tried removing some columns that I thought might not have as much of an impact on diabetes... but still accuracy is 70-80%.

[lazyprogrammer]

Here's a thought exercise. Can you construct a dataset in which it's not possible to obtain > 50% accuracy?

[lx0n2acl]

A dataset where the targets are all the same value?

[lazyprogrammer]

Try it and let us know!

[lx0n2acl]

So I am able to create a dataset where all of the training targets are 0, and 50% of the validation targets are 0 and the other 50% are 1... this results in a validation accuracy of 0.50.

[lx0n2acl]

Hey LP, how does this knowledge help to understand how I would go about getting a higher accuracy? I'm not seeing the connection... Unless the answer is, 76% is good enough for this particular dataset.