NLP based Industrial Accident Severity Assessment

• Domain: Industrial Safety

• Dataset: Industrial Safety and Health Analytics Database

• Context: The database originates from one of the largest industries in Brazil and globally. It is crucial for industries and companies worldwide to understand why employees continue to experience injuries and accidents in industrial settings, and in some tragic cases, fatalities. Addressing these issues is essential for improving workplace safety and preventing future incidents.

• Objective: To develop a machine learning/deep learning model that assists professionals in identifying and highlighting safety risks based on incident descriptions.

• Key Methodologies:

  • Data Analysis and Pre-processing
  • Vectorization: BoW, N-Grams, TF-IDF, Word2Vec, GloVe
  • Data Balancing: SMOTE, Text Augmentation using Llama3
  • ML for Classification: SVM, Ensemble Learning, ANN, BERT+LSTM