I am working on a text classification problem where I would like to improve the accuracy of my model. Presently, I am using SVM with linear SVC and OneVsRestClassifier. The model should correctly predict all of the subcategories for a parent category.

For entered account name, I should get the right description for my test data file that I will use to test the model later.

enter image description here

If I add another feature, +1 or -1, for debit and credit to my Account_Name field, will there any improvement in the accuracy?

Or, is there anything else that will help improve the accuracy of the model?

It is currently at 74% which is low.


It is hard to say without trying it. But you should add features as much as you can. When you have so many features, according to your success metrics, you can do future selection / regularization for both increasing the accuracy and speeding up the convergence of gradient descent.

Also you can also try (after adding more features) other classification algorithms such as RandomForest, GBM, Xgboost etc - Tree based models are more accurate in recent years.

Hope this is helpful!

| improve this answer | |
  • $\begingroup$ I have already tried RandomForest, Tree-based models, NB Classifier but the best result that I get is still from SVM. Could the accuracy be affected due to a combination of both English and German words in my columns? Like if in the first column if I have a combination of both English and German word bag like Software - Soll then will that affect the accuracy? $\endgroup$ – user79322 Aug 16 '19 at 9:59
  • $\begingroup$ Machine sees them as different things therefore their effect on convergence will be different even though they are same. In this case, I suggest you to translate german words to english and make them similar observations. $\endgroup$ – Ilker Kurtulus Sep 27 '19 at 9:25

Your Answer

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