1
$\begingroup$

I am very new to Data Science would appreciate your advice big time.

Got a task: predict if a trade will be profitable or not, based on a set of data. I have prepared, cleaned and tested data. Selected a few two-class algorithms since the answer requires a boolean result (true or false (1/0)). After testing selected two best from my opinion: Random Forest and Logistic Regression CV (cross validation). Below are the results. Can't decide which one to use in production. Also, please note: it is better not to trade than trade and loose money, therefore false negatives are better than false positives.

Random Forest results:

enter image description here

Logistic Regression CV (cross validation) results:

enter image description here

Added a sample of dataset for clarification:

enter image description here

Improve this question
$\endgroup$
1
$\begingroup$

There must be definitely some leakage in the data, which is why the Logistic Regression is predicting almost all as the positive class. You may want to check on that first before you proceed. The only reason Random Forest may seem better in that case is that it, by default, only takes a subset of variables.

Improve this answer
$\endgroup$
4
  • $\begingroup$ Could you please clarify what do you mean by leakage in the data? $\endgroup$ – Learning to program Dec 28 '18 at 18:15
  • $\begingroup$ Check the features more thoroughly to see if any information about the target feature is self contained in them. The logistic regression results definitely seems 'fishy' since everything is predicted as a single class. $\endgroup$ – lejin Dec 28 '18 at 19:36
  • $\begingroup$ thanks for your answer. Doesn't seem that the target feature is self contained in any of features. Added a screenshot of my data to my question above. $\endgroup$ – Learning to program Dec 28 '18 at 19:48
  • $\begingroup$ can you list the coefficients for the features in the trained log reg model? also, can you check predicting on your X_train and generate a confusion matrix on the training dataset? $\endgroup$ – lejin Dec 28 '18 at 22:00

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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