Need suggestions to improve accuracy for Text Classification Problem

Community
1

Hi Experts,

We are dealing with Text Classification Problem. We have around 80K records with around 50 classes. The data is highly imbalanced. It has 2 columns one for description and other contains class.

Till now we have tried following models and techniques:

  1. Data Preprocessing:
  2. Lowercase conversion, removed numeric texts, removed punctuations
  3. Removed unimportant words and stop words
  4. Lemmatization
  5. TFIDF transformation
  6. Using SKLEARN Models:
  7. Linear SVC
  8. Linear Regression
  9. Logistic Regression
  10. Decision Trees
  11. Random Forest
  12. Using Huggingface Transformers:
  13. Google Bert
  14. Distil Bert
  15. SMOTE sampling

It is observed that the maximum accuracy we got is 70% (Random Forest and Google Bert).

Is there any scope to improve accuracy?

If yes, what other techniques or models we can use to improve accuracy?