Air quality monitoring and analysis with dynamic training using deep learning
- PDF / 2,347,087 Bytes
- 20 Pages / 439.37 x 666.142 pts Page_size
- 5 Downloads / 173 Views
Air quality monitoring and analysis with dynamic training using deep learning Endah Kristiani1,2 · Ching‑Fang Lee3 · Chao‑Tung Yang3,4 · Chin‑Yin Huang1 · Yu‑Tse Tsan5,6 · Wei‑Cheng Chan5 Accepted: 23 October 2020 © Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract Time series prediction is a challenging predictive modeling case. It is essential to have a prediction model that can adapt to dynamic data. Air quality data show a significant changing degree of spatial and temporal data. Therefore, the updated deep learning model is suitable for this case. In this paper, monitoring and analysis of air quality with dynamic training using recurrent neural network (RNN) are proposed to provide the model remains up-to-date as new data comes. In the experiments, by adjusting the model, the accuracy is enhanced. The scheduling retrained model is provided based on the expected mean absolute percentage error (MAPE) value. First, the machine learning architecture environment is being prepared. Secondly, the RNN parameters were optimized for excellent level predictive precision. Third, set and test the scheduling and MAPE value based on the MAPE’s expected value for the automatic retraining model. Finally, on the interactive map, the output is presented using R and Shiny to visualize the RNN training results. Keywords Air pollution · PM2.5 · Time series · Deep learning · Dynamic forecasting · RNN
1 Introduction One primary consideration when developing a machine learning system is whether the training models are in a static, offline or dynamic way. Time series, also known as a series of data indexed in time order or dynamic sequences, is a challenging predictive modeling [5, 25]. Because of the dynamic and dependency values, it might end up with screwy predictions if the distribution of inputs changes and the model has not adapted. Therefore, to have a model that can evolve over time as new data comes in is essential, specifically in time series cases [23, 24]. * Chao‑Tung Yang [email protected] Extended author information available on the last page of the article
13
Vol.:(0123456789)
E. Kristiani et al.
Recurrent neural network (RNN) [7, 19] could manage the issue of time sequence as they can maintain state from one iteration to the next by using their output as input for the next phase. Their capacity to form a guided cycle has shown that RNN is one of a strong predictive engine, specifically for time series data. Air pollution dataset is a time series of records with natural temporal ordering and continuous records [15]. Thus, RNN is helpful to model and predict air pollution data in time series [17, 18]. In this paper, the monitoring and analysis of air quality with dynamic training using recurrent neural network (RNN) is proposed. In the experiments, the accuracy was enhanced by adjusting the model and provide scheduling retrained model in an expected mean absolute percentage error (MAPE) value. First, the environment for the machine learning process is prepared [13]. Second, the RNN pa
Data Loading...
