Comparative Analysis of Regression-Based Forecasting Models for State-Level E-Waste Generation Prediction in India

Abstract

Electrical and electronic waste (or e-waste) constitute one of the most rapidly growing waste streams in India but, unfortunately, the accuracy of forecasting of state-wise generation of e-waste based on statistical analysis is seldom verified using real observations made after forecasting. In this research, two regression-based forecasting techniques, namely simple linear regression and second degree polynomial regression, are compared as forecasting methods for state-wise prediction of e-waste generation in India in the year 2024 based on historical data available with the Central Pollution Control Board (CPCB) from the years 2016 through 2023. The predictions have been validated using actual observations made in 2024 rather than goodness of fit, which makes this paper unique in comparison with many forecasting studies available. Polynomial regression resulted in slightly better mean absolute error (14,998 metric tonnes as opposed to 16,993 metric tonnes in linear regression case), and Pearson correlation of r = 0.893 as compared with r = 0.914 in the case of the linear regression method, although the linear regression technique produced slightly lower root-mean-square error and higher coefficient of determination, which was attributed to a single outlier resulting from polynomial regression for Uttarakhand. Accuracy at the state-level was quite different from the state to the state: three states - Tamil Nadu, Uttarakhand, and Maharashtra - were validated within five per cent error with the actual 2024 value, whereas the three other states, namely Andhra Pradesh, Himachal Pradesh, and Chhattisgarh, had forecasting error rates above 150%. These discrepancies seem to correlate more with the internal consistence of data series provided by particular states rather than with the used forecasting technique. It may be summarized that forecasting accuracy in Indian e-waste industry does not depend on mathematical complexity of used technique but more on the consistency of historical data provided. Thus, the most important outcome of this research is the following recommendation for State Pollution Control Boards to improve data-reporting processes rather than implement sophisticated predictive techniques