A Comprehensive Evaluation of Machine Learning Models for Sentiment Analysis in Employee Reviews
Ananto Tri Sasongko, Muhamad Ekhsan, Wahyu Hadikristanto, Agung Nugroho · 2024
In natural language processing, sentiment analysis is critical in interpreting textual data to extract subjective information. This study aims to comprehensively evaluate various machine learning models for sentiment analysis, specifically applied to employee reviews. Utilizing LazyPredict, we compare the performance of multiple classifiers to identify the most effective ones for this task. The dataset comprises employee reviews, including positive (Pros) and negative (Cons) feedback and overall ratings. We incorporate additional textual attributes such as text length and word count to enhance the feature set for both Pros and Cons. The data undergoes preprocessing, followed by handling class imbalance through SMOTE and dimensionality reduction via PCA. Experimental results highlight each model's accuracy, precision, recall, and Fl-score, providing a detailed comparative analysis. The study reveals that models such as Random Forest consistently outperform others regarding overall effectiveness in sentiment classification, scoring 94% for all metrics. The findings are significant for organizations seeking to leverage employee feedback for strategic decision- making. By identifying the most accurate and reliable models for sentiment analysis, businesses can better understand employee sentiment, leading to improved employee engagement, satisfaction, and retention strategies.