Evaluating Relative Efficiency of ML Models for Classification of Disaster Tweets using NLP
disastAIr is a student project as part of the Machine Learning course at Vrije Universiteit Amsterdam. The project encompasses training and evaluation of two different ML models for classification of disaster related Tweets using NLP.
While the project is based on a Kaggle competition, the aim of the project goes beyond that of the competition. Namely, the project sets a focus on evaluating different ML models not only based on their performance but also on their footprint cost.
The models used are:
- a classical model in the form of a Markov classifier, and
- a deep learning (DL) model in the form of a distilled version of the transformer-based BERT model (DistilBERT).
Python is used for all models and experiments. The DL model is used through the KerasNLP library, while the Markov classifier is implemented fully from scratch.
A full project description, including results and a discussion, is given in the project report.
| Name | Profile |
|---|---|
| Maria P. Jimenez M. | GitHub, LinkedIn |
| Lennart K.M. Schulz | GitHub, LinkedIn |
| Laura I.M. Stampf | GitHub, LinkedIn |
| Dovydas Vadišius | GitHub, LinkedIn |