Multi-site household waste generation forecasting using a deep learning approach

Area of Science:

• Environmental Science
• Data Science
• Machine Learning

Background:

• Traditional methods for forecasting household waste generation face challenges due to high variability and non-linear dynamics.
• Existing studies often focus on municipal or country levels, overlooking the rapid short-term variations in household data.

Purpose of the Study:

• To evaluate the effectiveness of a state-of-the-art deep learning approach for household waste generation forecasting.
• To compare the performance of a Long Short-Term Memory (LSTM) neural network against traditional forecasting methods.

Main Methods:

• Application of a multi-site Long Short-Term Memory (LSTM) Neural Network model.
• Utilizing a long-term database of weekly household waste weights from Herning, Denmark (2011-2018).
• Comparison with traditional time-series models like ARIMA.

Main Results:

• The LSTM model demonstrated a significant improvement in forecasting accuracy compared to traditional methods, achieving an average improvement of 85%.
• A multi-site approach within the LSTM model enhanced forecasting performance by an average of 28% compared to individual household fits.
• The deep learning approach effectively captured the complex, non-linear dynamics of household waste generation.

Conclusions:

• Deep learning, specifically the multi-site LSTM approach, offers a superior method for forecasting household waste generation compared to traditional techniques.
• The findings highlight the potential of advanced machine learning models in managing waste streams more effectively.
• Accurate waste generation forecasting can support better resource allocation and waste management strategies.