Feature engineering is a crucial step for natural language processing (NLP) applications, involving the extraction of pertinent features from unprocessed text data and their subsequent conversion into a format suitable for machine learning models. In this study, we compare feature engineering techniques using two datasets. These datasets concern the medical domain, where we applied widely used feature engineering techniques to convert textual documents into numerical vectors representing textual features. The three techniques TF-IDF, GloVe and Word2Vec were used to be applied to both cancer text documents and Symptom2Disease datasets. Then, we trained some classification models and evaluated the performance of each model using standard evaluation metrics. The results indicate that the performance differs depending on the used feature extraction techniques. The LR model with TF-IDF features achieved the highest accuracy for the first dataset compared to the other models, and the KNN model using TF-IDF features obtained the best accuracy for the second dataset. Furthermore, these results demonstrate that the TF-IDF features outperform GloVe and Word2Vec.