System of ecological monitoring of the environment using neural networks and time series
Students Name: Sulymka Roman
Qualification Level: magister
Speciality: Software Engineering
Institute: Institute of Computer Science and Information Technologies
Mode of Study: full
Academic Year: 2022-2023 н.р.
Language of Defence: ukrainian
Abstract: Modern methods of pollution forecasting are considered, their advantages and disadvantages are analyzed. The article demonstrates the result of predicting soil contamination using forecasting models based on ARIMA and LSTM time series. A set of input data was analyzed to eliminate missing values of soil contamination samples. Models for forecasting time series were analyzed: naive method, ARIMA, SVR and LSTM. The pros and cons of each of the analyzed models were determined, and as a result of the analysis, two models were chosen for the study — ARIMA and LSTM. Since the LSTM model belongs to neural networks, transfer learning and fine-tuning techniques are applied to it to adapt the model to the problem of soil pollution prediction. It has been established that the use of the above techniques allows obtaining high accuracy rates of pollution forecasting using a neural network. A comparative analysis of soil pollution forecasting using these models was carried out. An air pollution forecast based on ARIMA and LSTM models was implemented using a data set of one thousand records, in order to compare how the model is affected by the amount of input data. The results of the conducted research indicate the high accuracy of the ARIMA algorithm, and the use of the LSTM model for solving the forecasting problem is recommended only with a large sample of input data and the correct setting of its parameters. The software implementation of the models is developed in the Python programming language using such libraries as: Numpy - for the implementation of forecasting algorithms, Pandas for processing input data and Seaborn for visualization of time series and the results of forecasted values. Based on the analyzed parameters, the ARIMA model was chosen as the optimal one for predicting soil contamination. Keywords: Time series, soil pollution, pollution forecasting, ARIMA, LSTM.