Forecasting of Municipal Waste Generation Using Machine Learning

Effective waste management is crucial for tackling various environmental challenges, where precise predictions of waste generation play a vital role in planning and decision-making. Traditional prediction methods often rely on limited historical data and simplistic models, resulting in potential inaccuracies and unreliable forecasts. Artificial intelligence offers a groundbreaking approach to enhance the accuracy and reliability of waste generation predictions. Unlike traditional models, AI can process and analyze vast amounts of data from diverse sources, utilizing sophisticated machine learning algorithms to make more informed and accurate predictions. We used global autoregressive models with sequential (event-time) lags, which better capture irregular collections (collection three times per week). The evaluation was performed on a joint test with an 80/20 stand split. The best performance was achieved by LightGBM-Tweedie with an RMSE of 108.4 kg, which means an improvement of 41.5% compared to the naive baseline (185.3 kg) and 19.8% compared to the baseline model based on previous values on the same day of the week (135.2 kg).