Article
A Transformer-Guided Adaptive Model for Workforce Satisfaction Prediction
The expansion of digital platforms has resulted in a substantial increase in employee-generated feedback, making the assessment of workforce satisfaction a critical research focus. Conventional methods, such as manual surveys and basic statistical analyses, are often inefficient and inadequate for extracting meaningful insights from complex textual data. Recent developments in Natural Language Processing (NLP) and Machine Learning (ML) have enabled automated approaches; however, many existing solutions face challenges with unstructured text, imbalanced datasets, and multi-label prediction tasks. This study focuses on improving the prediction of key workforce satisfaction dimensions, including work-life balance, skill enhancement, compensation and benefits, job stability, career advancement, and overall job satisfaction, based on employee reviews. Traditional techniques are limited in scalability and often deliver inconsistent results, highlighting the necessity for a more advanced and reliable analytical framework. To address these limitations, the proposed approach combines NLP-based preprocessing, transformer-driven feature extraction using Google PaLM (Pathways Language Model), and the Synthetic Minority Over-sampling Technique (SMOTE). It evaluates multiple machine learning models Quadratic Discriminant Analysis (QDA), Linear Discriminant Analysis (LDA), and Histogram-Based Gradient Boosting (HGB) and compares them with a newly introduced Transformer-Guided Adaptive Model (TGAM). Experimental findings indicate that conventional models achieve accuracy levels between approximately 51% and 56%, whereas the TGAM model attains a perfect accuracy of 100% across all evaluated categories. These results demonstrate the robustness and effectiveness of the proposed system in analyzing complex employee feedback data, supported by comprehensive evaluation metrics and visualization methods for enhanced interpretability.
Full Text Attachment
