A Hybrid Type-۲ Fuzzy-LSTM Model for Prediction of Environmental Temporal Patterns

Computational intelligence methods, such as fuzzy logic and deep neural networks, are robust models to solve real-world problems. In many dynamic and complex problems, statistical attributes frequently change over the time. Recurrent neural networks (RNN) are suitable to model dynamic high-dimensional and non-linear state-space systems. Nevertheless, the RNN is incapable of modelling long-term dependencies in temporal data, and its learning using gradient descent is a complex and difficult task. Long Short-Term Memory (LSTM) networks were introduced to overcome the RNN issues, but coping with uncertainty is still a major challenge for the LSTM models. This research presents a Hybrid Type-۲ Fuzzy LSTM (HHT۲FLSTM) deep approach to learn long-term dependencies in order to obtain a reliable prediction in uncertain time series circumstances. The proposed model was applied to the air quality prediction problem to evaluate the model’s robustness in handling uncertainties in a real-world application. The proposed model has been evaluated on a real dataset that contains the outdoor pollutants from July ۲۰۱۱ to October ۲۰۲۰ in Tehran and Beijing by a ۱۰-fold cv with an average area under the ROC curve of ۹۷ % with a ۹۵% confidence interval [۹۵-۹۷] %.

Keywords:

deep learning , Type-۲ Fuzzy Logic , LSTM network , Time-Series Prediction