Determinants of electronic waste generation in Bitcoin network : Evidence from the machine learning approach

Abstract

Electronic waste is generating in the Bitcoin network at an alarming rate. This study identifies the determinants of electronic waste generation in the Bitcoin network using machine learning algorithms. We model the evolutionary patterns of electronic waste and carry out a predictive analytics exercise to achieve this objective. The Maximal Information Coefficient (MIC) and Generalized Mean Information Coefficient (GMIC) help to study the association structure. A series of six state-of-the-art machine learning algorithms - Gradient Boosting (GB), Regularized Random Forest (RRF), Bagging-Multiple Adaptive Regression Splines (BM), Hybrid Neuro Fuzzy Inference Systems (HYFIS), Self-Organizing Map (SOM), and Quantile Regression Neural Network (QRNN) are used separately for predictive modeling. We compare the predictive performance of all the algorithms. Statistically, the GB is a superior model followed by RRF. The performance of SOM is the least accurate. Our findings reveal that the blockchain's size, energy consumption, and the historical number of Bitcoin are the most determinants of electronic waste generation in the Bitcoin network. The overall findings bring out exciting insights into practical relevance for effectively curbing electronic waste accumulation.