Towards accurate and efficient waste image classification: A hybrid deep learning and machine learning approach

Automated image-based garbage classification is a critical component of global waste management; however, systematic benchmarks integrating machine learning (ML), deep learning (DL), and efficient hybrid solutions remain underdeveloped. This study provides a comprehensive comparison across three paradigms: (1) ML algorithms using handcrafted features, (2) DL architectures, including ResNet variants, EfficientNetV2S, DenseNet121 and MobileNetV3, and (3) a hybrid approach using deep models for feature extraction combined with various ML classifiers to identify the most effective strategy. Experiments on three public datasets demonstrate that the hybrid method consistently outperforms the others, achieving up to 99.72% accuracy on TrashNet, 100% on Household, and 99.87% on Garbage dataset, surpassing state-of-the-art benchmarks. Furthermore, feature selection reduces feature dimensionality by over 95% without compromising accuracy, resulting in faster training and inference. This work establishes more reliable benchmarks and introduces an efficient hybrid framework, achieving high accuracy while reducing inference cost, making it suitable for scalable deployment. • Systematically evaluates ML, DL, and hybrid approaches for waste image classification. • Refines the Household Garbage dataset by correcting 43 mislabeled samples. • Applies feature selection to reduce dimensionality by over 95% without any loss. • Proposes a hybrid DL–ML framework achieving near-perfect accuracy and high efficiency.