Publication

Abstract : According to reports by Safe Food Advocacy Europe 2024, 220 million tons of plastic waste are disposed of as waste. Among them, 69.5 million tons end up in the environment, which poses hazards to humans and aquatic organisms. Pyrolysis is a thermochemical recycling method that converts plastics into liquid, solid, and gaseous products in an inert atmosphere at 400-600 oC. The objective of the study is to predict the product yield using various machine learning algorithms (linear regression, decision trees, random forest, gradient boosting machines and support vector machines). The challenges with plastic pyrolysis industries are the need for capital investments and operating expenditure, difficulties in pilot-scale experiments and optimization of product quality due to diversity in the composition of plastics. In comparison with the literature, the present work incorporates a variety of feedstocks such as lowdensity polyethylene (LDPE), polypropylene (PP), acrylonitrile butadiene styrene (ABS), high density polyethylene (HDPE), polycarbonate (PC) and polystyrene (PS) to predict the product yield and quality from plastic pyrolysis. The methodology comprises data collection from published literature, processing of numerical parameters (temperature and pressure), and categorical parameters (feedstock, catalyst, reactor type). The product outputs (solid, liquid, and gas) are correlated with published findings, ensuring that they fall within realistic ranges. Solid yields are typically observed between 5% and 15%, liquid yields range from 60% to 90%, and gas yields are calculated to ensure the total sums to 100%. Furthermore, the dataset is split into train and test sets in the ratio of 80:20 followed by handling outliers using isolation forest, data visualization, and evaluation of best models for the yield prediction. The preliminary data analysis showed that the random forest model showed a satisfactory R-square score of 0.92 compared to other models due to its ensemble behaviour. Temperature was found to be the dominant parameter in influencing liquid and gas yield. The heat correlation map showed that carbon content has a strong positive correlation with solid yield and a negative correlation with gas yield. The prediction can be improved by using a large volume of data sets, incorporating new machine learning algorithms and the testing with pilot scale data sets. The graphical user interface for plastic pyrolysis would be a better option for pyrolysis industries to optimize product quality and yield from a variety of feedstocks.