Power Outage Detection

This GitHub repository contains code for power outage detection using transfer learning and language models. The code explores the application of pretraining and fine-tuning language models to detect power outages, even with limited labeled data. It also includes baselines using classical models for comparison.

Data

The repository provides code to load and preprocess the data. It uses a CSV file (data.csv) containing social media tweets associated with power outages. The data is split into training and testing sets, ensuring a balanced distribution of samples for each class ('outage' and 'no_outage').

Baselines

The repository includes implementation of baseline models using classical machine learning algorithms. Specifically, it demonstrates the use of TF-IDF vectorization and classification models such as XGBoost, SVM, and Logistic Regression. The evaluation metrics, including accuracy, precision, recall, and F1-score, are calculated for each baseline model.

Transfer Learning with Language Models (LLMs)

The code showcases the application of transfer learning with language models for power outage detection. It includes zero-shot learning and few-shot learning approaches using popular LLMs like BERT and GPT. The evaluation metrics are calculated for each LLM and finetuning percentage.

Zero-Shot Learning

The repository provides code to perform zero-shot classification using the BERT and GPT language models. It uses a pipeline approach to classify the testing data into 'outage' and 'no_outage' categories. Evaluation metrics such as accuracy, precision, recall, and F1-score are calculated.

Few-Shot Learning

The code demonstrates few-shot learning by fine-tuning the language models with a limited percentage of training data. It randomly samples a balanced subset of the training data and evaluates the performance on the testing data. The evaluation metrics, including accuracy, precision, recall, and F1-score, are calculated for each finetuning percentage and model type (BERT and GPT).

Prerequisites

To run the code in this repository, ensure that you have the following dependencies installed:

• pandas
• torch
• scikit-learn
• transformers
• xgboost