Initially, remove the transcriptions having category labels less

than $50$ in the corpus as in $[2].$ Apply data preprocessing techniques also following the steps in $[2].$ For

feature extraction, apply Bag$-$of$-$Words $($CountVectorizer$)$ and TF$-$IDF $($TfidfVectorizer$)$ separately.

Implement Multinomial Na$$ve Bayes, Random Forest, XGBoost, LightGBM for the traditional machine

learning algorithms of the medical text classification process. Then, apply at least one complex deep

neural network architecture $($ensemble learning$)$ using $1$D CNN$,$ LSTM and GRU. Show the confusion

matrix, accuracy, precision, recall and F$1-$score for each category class of the implemented solutions.

In the next phase, use the NER code previously implemented in the first part of the project. Use the

labeled named entities and their category labels as the input, then follow the same training and

evaluation steps.

Finally, apply SMOTE oversampling method $[2]$ for the best accuracy values in the previous two phases

and compare accuracy, precision, recall and F$1-$score with and without oversampling. Write a report that

explains and illustrates the results step by step