Rapid urbanization has created a critical global challenge in waste management. The significant increase in waste generation is driven by changing consumption habits, inadequate waste sorting, and the widespread use of single-use plastics. This work investigated the application of data-driven models and image analysis for automated waste detection and classification. Specifically, we trained and evaluated the YOLOv8 deep learning model using a combination of locally sourced waste dataset and a publicly available dataset of beverage waste. A key focus was to ensure model stability and generalizability across diverse data sources. Integrating a partial portion of open data into the training process led to a slight improvement in performance on the internal test set, while the external test set demonstrated a more substantial performance gain compared to training with locally sourced dataset only. Overall, our findings demonstrated the promising potential of automated waste segregation systems. Further enhancements, such as expanding dataset diversity and refining model architectures, could contribute to more efficient and sustainable waste management practices.