Prediction of e-waste generation: Application of modified adaptive neuro-fuzzy inference system (MANFIS)

An accurate estimation of generated electronic waste (e-waste) plays a pivotal role in the development of any appropriate e-waste management plan. The present study aimed to exploit modified adaptive neuro-fuzzy inference system (MANFIS) for the estimation of generated e-waste. There are different parameters affecting e-waste generation, the most important of which need to be identified to achieve the accurate estimation. The MANFIS used for parameter selection involves evaluating multiple choices between twelve initially specified parameters. The MANFIS models with five inputs have the highest mean R²_(train) and R²_(test) (0.978 and 0.952, respectively, in training and testing stages). According to the results, the best combination of parameters was related to legal imports of electrical and electronic equipment (EEE), smuggling (illegal) imports of EEE, exports of EEE, accumulation of EEE in Tehran, and accumulation of EEE in Iran with RMSE_(train) and RMSE_(test) of 0.221 and 2.221, respectively. The findings showed that the model with three triangular membership functions had the best performance; R²_(train) and RMSE_(train) values were 0.981 and 1.371, as well as R²_(test) and RMSE_(test) values were 0.971 and 1.678, respectively. Finally, the developed model was successfully applied for prediction of monthly e-waste generation in Tehran for thirteen selected electronic items. The obtained consistent results emphasized that appropriate selection of the number of input parameters and their combination, along with identifying optimal structure of MANFIS, provides a proper, simple and accurate prediction of e-waste.