email-classification-models

Project objective

The goal of this project is to build a model which classifies emails as ham or spam

Project description

• A dataset of 5574 emalis and 2 features (text, label) is given
• The text was cleaned by removing punctuation and stopwords
• The clean text was vectorized using TfidfVectorizer
• Classic models were built such as: LogR, RF, GB
• Accuracy, Precision and Recall measures were calculated
• Resampling technique was applied to balance the data
• Basic RNN model (sequential) was built
• Dimensionality reduction was executed using PCA method

Conclusions

• The highest scores were achieved for Random forest model with: acc = 0.976, pre = 0.972, rec = 1
• The results got improved a bit using the resampling method, but computation time was increased
• Models with high acc\pre but low rec are not relevant for this type of project
• The fastest model was Logistic regression with very good results: acc = 0.95, pre = 0.95, rec = 0.995
• Using PCA method to reduce dimensionality we got very nice results and less overfitting problem

The most accurate model for this dataset was RF, and the fastest was LogR