A Machine Learning Approach to Indonesian Climate Change Sentiment Analysis Using Naive Bayes

Abstract

Climate change poses a significant global challenge, particularly for archipelagic nations such as Indonesia that are highly vulnerable to rising temperatures and extreme weather events. This study applies machine learning-based sentiment analysis to assess Indonesian public opinion on climate change using Twitter data. A total of 5,120 Indonesian-language tweets were collected through keyword-based scraping related to climate and weather conditions. Following text preprocessing (lowercasing, stopword removal, stemming, and cleaning), TF-IDF vectorization was used to extract the top 400 most significant terms. The dataset was divided into training (80%) and testing (20%) subsets, and a Multinomial Naïve Bayes classifier was trained to categorize sentiments into positive, neutral, and negative classes. The results show a dominance of negative sentiment (62%), primarily associated with extreme heat and storm-related events, while neutral (24%) and positive (14%) sentiments were linked to moderate weather conditions. Model evaluation achieved an F1-score of 0.95 for negative, 0.86 for neutral, and 0.83 for positive sentiment, yielding a macro-average F1-score of 0.88. The analysis also identified “panas (hot),” “hujan (rain),” and “banjir (flood)” as top lexical indicators influencing classification. Overall, the findings highlight that Indonesian public sentiment toward climate change is highly reactive to extreme weather. The study underscores the potential of Naïve Bayes as a baseline model for real-time environmental sentiment monitoring, offering valuable insights for institutions such as BMKG to enhance public communication and climate awareness strategies.