I was wondering how to interpret and tweak a model with 10% accuracy, in binary classification. With common feature engineering, I get something around 65% accuracy. When I do some transformation on the features( fft) it goes down to around 10%. This tells that if I flip my predictions it would be 90% accuracy. As an example in stock exchange when it says buy I sell and vice versa. very Poor binary classification is very good when I reverse it. My question is that, what can I do to in fact make the model look reasonable. Or what is the next CORRECT step in this situation, toward improving the model? It just don't feel right to flip the predictions.
Asked
Active
Viewed 14 times
0
-
If your accuracy in a binary task is 10%, something has gone badly wrong in your model. I don't think we can diagnose the problem without knowing *a lot* more about your data, your model etc. Also: [Why is accuracy not the best measure for assessing classification models?](https://stats.stackexchange.com/q/312780/1352) – Stephan Kolassa Nov 12 '21 at 17:05
-
Thanks @StephanKolassa. I actually measure all the scores, and all are bad(good). I agree that something has gone wrong, but not very bad, other wise the accuracy would be 50%, right? I made sure its is not random, by changing random_state, many times, – Nabat Farsi Nov 12 '21 at 17:14