Development of Forecasting Model for Machine Learning-Based Landfill Leachate Generation Using Linear Interpolation

Abstract

Purpose : The purpose of this study is to compare single models according to the missing value handling techniques for predicting leachate generation. Method : Input factors such as leachate generation, landfill gas generation, and weather data (precipitation, wind speed, radiation, temperature, and relative humidity) were used from June 2002 to October 2018. Linear interpolation and mean method were used as the missing value handling technique. The experiment was conducted by dividing the data into train and test data according to the optimal ratio. Various single models were used, and the prediction performance of the model was compared and evaluated using coefficient of determination (R2). Result and discussion : The gated recurrent unit (GRU) model was the best among the single models. In the GRU model, R2 was 0.867 for linear interpolation and R2 was 0.839 for the mean method, so that the GRU model using linear interpolation was most suitable for predicting leachate generation. In the ANN model, R2 of linear interpolation (0.862) was higher than that of mean method (0.828). In the long short-term memory (LSTM) model, R2 was 0.779 for linear interpolation and 0.762 for mean method. In the random forest (RF) model, R2 for linear interpolation (0.700) was also higher than that for the mean method (0.665). The model performance was excellent in the order GRU > ANN > LSTM > RF. The linear interpolation for the missing value handling technique was superior to the mean method in all models used in this experiment. Conclusion : The GRU using linear interpolation was the most suitable model for predicting landfill leachate generation.

Keywords

• gated recurrent unit
• landfill
• leachate
• linear interpolation
• machine learning