Water quality prediction based on AR and LSTM model

Document Type

Conference Proceeding

Publication Date



Prediction for water quality is essential because poor water quality can result in serious health problems. Turbidity is one indicator to measure water quality. This research uses time series models, the Autoregressive (AR) model and the Long short-term memory (LSTM) model, to predict the turbidity of water based on the dataset collected from Chicago Park District with a long-term time series from 2013 to 2021. Data cleaning and data exploration are done before building the model, and the stationary test and seasonality test are executed to prepare qualified time series. Also, ACF and PACF plots are drawn to figure out the order of AR to select the best model for prediction. AR Model is a traditional time series model which can make a prediction based on the previous values of time series, and LSTM Model is an advanced Recurrent Neural Network (RNN) designed to prevent the gradient from either decaying or exploding when learning from long-term sequences. The research applies these two models to the water quality data and makes an evaluation and comparison between them. It is found that the LSTM Model has a better performance in forecasting a long-period time series than the AR Model.

Publication Title

Journal of Physics: Conference Series



This document is currently not available here.