Learning Models for Software Feature Extraction from Disaster Tweets: A Comparative Study

This study assesses the effectiveness of different combinations of word embeddings and machine learning models in classifying disaster-related tweets to improve emergency response efforts. The primary objective is to identify the most reliable approach for categorizing tweets as either disaster-related or unrelated. The research employs TF-IDF for feature vectorization and Word2Vec for word embedding, combined with machine learning models such as Logistic Regression, Support Vector Machines, K-Nearest Neighbors, Decision Trees, and Random Forest. After extensive preprocessing, the results indicate that Word2Vec combined with Logistic Regression achieved the highest performance, with 81% accuracy, precision, and recall, and an F1 score of 80%. Other combinations, including Word2Vec with Support Vector Machines and Random Forest, also demonstrated strong results. In contrast, TF-IDF combined with K-Nearest Neighbors resulted in the lowest accuracy at 68%. These findings highlight the critical importance of selecting the appropriate word embedding techniques and machine learning models for effective text classification. Future research should explore more advanced embeddings like BERT and Transformer, while also incorporating temporal and semantic analysis to further enhance classification accuracy and robustness.