A Novel Stack Ensemble Learning Techniques for Sentiment Classification in Climate Change Discourse

Primary reason of climate reason is release of greenhouse gases into the biosphere which warms the earth’s atmosphere. Climate change is serious issue that affects public sentiment significantly. In this paper, stack ensemble learning based framework is proposed to analyze public sentiment regarding climate change by using text datasets. First step of proposed framework is text preprocessing which includes tokenization, stemming, and stop-word removal to create TF-IDF representations. Next, stack ensemble model is developed for sentiment classification as positive, neutral, or negative. This model utilizes four machine such as support vector machine (SVM), decision tree (DT), random forests (RF), and gradient boosting classifier (GDB) as base models. The predictions given by all four models are combined using XGBoost as a meta-classifier. The four dataset related to climate change are used for experimental analysis. Results show that the proposed model achieves 89.72% precision, 91.65% recall, 90.23% F1-score, and 91.47% accuracy. These results suggest that the stack ensemble learning approach is effective for analyzing public sentiment about climate change.