Enhancing e-waste management: a novel light gradient AdaBoost support vector classification approach.
Journal:
Environmental monitoring and assessment
PMID:
40014109
Abstract
The global consequences of electronic waste significantly affect the environment and human health. Accurate classification is essential for effective recycling and management to mitigate serious environmental harm caused by improper disposal. However, the traditional approaches find it difficult to ensure the entire lifecycle of electronic products, which includes the design, production, and consumption stages. Therefore, this study introduces a novel light gradient AdaBoost support vector-based dendritic growth search algorithm for enhancing e-waste management. This approach aims to improve the accuracy of e-waste classification and streamline waste collection planning. Data collection, data preprocessing, feature extraction, and classification phase are the four primary modules in this framework. The process initiates with the data collection module where the data is obtained from three diverse datasets: the e-waste dataset, the CTSOC E-waste dataset, and the e-waste vision dataset. Then, the preprocessing module allows the acquired data to enhance the quality of images through several preprocessing steps, including image scaling, image rotation, flipping, noise removal, and label encoding. Following this, the feature extraction module extracts significant characteristics using modified principal component analysis. The proposed method employs a Light Gradient Boosting Machine for the accurate processing of e-waste features. AdaBoost is employed to identify relevant features and integrate multiple weak learners. Support vector machine is implemented to navigate high-dimensional spaces and reduce overfitting, while linear regression aids in predicting e-waste categories. Parameter tuning is optimized through dendritic growth search to enhance the classification process. The experimental results proved that the proposed model outperformed the existing models with an impressive precision of 98.1%, recall of 96.1%, F1-score of 97.1%, accuracy of 98.5%, and specificity of 97.3%, respectively. Overall, the proposed framework exhibits significant potential for optimizing e-waste management.