Feature Selection using Particle Swarm Optimization for Sentiment Analysis of Drug Reviews

Abstract

Feature selection (FS) is an essential classification pre-processing task that eliminates irrelevant, redundant, and noisy features. The primary benefits of performing this task include enhanced model performance, reduced computational expense, and modified “curse of dimensionality”. The goal of performing FS is to find the best feature group that can be used to build an effective pattern recognition model. Drug reviews play a significant role in delivering valuable medical care information, such as the efficacy, side effects, and symptoms of drug use, facilities, drug pricing, and personal drug usage experience to healthcare providers and patients. FS can be used to obtain relevant and valuable information that can produce an optimal subset of features to help obtain accurate results in the classification of drug reviews. The FS approach reduces the number of input variables by eliminating redundant or irrelevant features and narrowing the collection of features to those most significant to the machine learning model. However, the high dimensionality of the feature vector is a major issue that reduces the accuracy of sentiment classification, making it challenging to find the best feature subset. Thus, this article presents a perceptive method to perform FS by gathering information from the potential solutions generated by a particle swarm optimization (PSO) algorithm. This research aimed to apply this algorithm to identify the optimal feature subset of drug reviews to improve the classification accuracy of sentiment analysis. The experimental results showed that PSO provided a better classification performance than a genetic algorithm (GA) and ant colony optimization (ACO) in most datasets. The results showed that PSO demonstrated the highest levels of performance, with an average of 49.3% for precision, 73.6% for recall, 59% for F-score, and 57.2% for accuracy.