Feature Selection Technique Model for Forest and Land Fire Data Sentiment Analysis: Comparison of SSA, PSO, and ALO

The forest and land fires in Indonesia caused a trending topic on social media, namely Twitter, especially in 2019 with 316,782 tweets. Mining it with a sentiment analysis approach will yield valuable knowledge. Mining data from Twitter that contain huge quantities cannot be handled with a sentiment analysis approach using conventional machine learning, where the data are very rich in features. This research aims to select the most relevant and informative features using the feature selection technique (FST), which adopts a Swarm Intelligence-based algorithm to produce accurate and fast model performance in processing time. This study involves several phases, including data collection, feature selection, labeling, data splitting, text pre-processing, modeling, and performance assessment. Salp Swarm Algorithm (SSA), Particle Swarm Optimization (PSO), and Ant Lion Optimization (ALO) will be compared in this research as feature selection technique optimization using classification algorithms such as K-Nearest Neighbor (KNN), Support Vector Machine (SVM), and Naive Bayes (NB). Research results based on algorithm performance evaluation show that the KNN Algorithm using the PSO algorithm is the algorithm that performs best, with an accuracy of 90.17% and the fastest processing time, namely 0.038 seconds. KNN Algorithm using the SSA algorithm is also no less superior, achieving accuracy reaching 90.02% with a processing time of 0.039 seconds.