README.md

This contains EDA and pipeline formation for cyberbullying dataset (youtub_parserd_dataset)

To access the pipeline go to pipeline.ipynb to test the pipeline on other dataset for "youtube_parsed_dataset".

Changes

Another file FinalPipeline2.ipynb is made in which we used numerical_columns as test parameters instead of processed_text so as to evaluate Accuracy and AUC scores on it.

FinalPipeline_Report.ipynb and FinalPipeline2_Report.ipynb are added in which I tried to provide a good explaination of code and its output for both FinalPipeline.ipynb and FinalPipeline2.ipynb respectively.

Test Resuts Random forest : Achieved AUC score of around 0.99 and outperformed every other model in both cases. Logistic Regression: Achieved an accuracy of approximately 91.9% and an AUC score of about 97.1%, indicating good per formance in distinguishing between cyberbullying and non-bullying instances. XGBoost: Outperformed logistic regression with higher accuracy (~94.8%) and AUC score (~98.8%), suggesting superior predictive capabilities. Decision Tree: Although providing acceptable performance with an accuracy of around 90.1% and an AUC score of about 90.4%, it lags behind logistic regression and XGBoost.

Change 2

LSTM model is integrated in a pipeline. The function is saved inside LSTM_pipeline.ipynb. AUC of 96.5 is obtained using LSTM model. Code description is also provided in the end.