Aspect-Based Sentiment Analysis (ABSA) has seen significant advances through the application of a wide range of deep learning techniques. Early approaches often employed convolutional neural networks (CNNs) and recurrent models such as Long Short-Term Memory (LSTM), and augmented with bi-directional structures to enhance contextual representation. Subsequent developments introduced attention mechanisms, which significantly improved the ability to model aspect-specific sentiment dependencies. More recently, the field has shifted toward transformer-based and large language models (LLMs), which provide strong contextual representations adaptable to ABSA tasks across domains and languages. Despite their success, these models often demand substantial computational resources and complex tuning. This study explores the use of gradient boosting for aspect-based sentiment analysis (ABSA) using a two-stage approach: aspect detection followed by sentiment classification. XGBoost was applied with basic feature engineering and preprocessing, and the model achieves a macro-averaged F1-score of 0.75 for aspect detection and 0.74 for sentiment classification on Vietnamese benchmark dataset UIT-ViSFD. The results suggest that XGBoost can be a useful and efficient option for ABSA, particularly in situations where computing resources are limited.