Fuzzy Ranking Algorithm with Seagull Optimization-Based Decision Tree for Short-Term/Long-Term Rainfall Prediction

Introduction

Worldwide weather forecasting is vital and challenges operational responsibilities recognized by meteorological facilities. Prediction is used as a source to raise awareness that is essential for an upcoming event. The ability to divert, store, and recycle (Maurya, S. P & Singh, 2021) water makes it one of the most manageable natural resources. To improve watershed planning and management, more efficient techniques are needed. Predicting rainfall is an essential and significant method today. Each year, heavy rains and floods kill and evacuate many people. Hydraulic analysis and modelling are necessary components of water resource application, and improving the accuracy of variability estimation in accounting water is mandatory (Wang, H. R et al., 2012). The precipitation process is associated with precursors from sub-regulators such as long-term surface pressure, sea surface temperature and other atmospheric parameters for seasonal time measurements rather than weekly and daily time scales. Limited additional atmospheric controls adapt to temperature, humidity, dew point and wind. During daily rainfall events, the unpredictability of weather and climatic conditions will be a major factor. If the unpredictable shape is documented and applied to the future path, the possibility of daily rainfall is very high (Edvin & Yudha ,2008). Rainfall has an impact on many human activities including agriculture, construction, power production, forestry, and tourism (McMichael, et al., 2003) . In most hydraulic and climatic models, rainfall is a required input variable; its precise accounting may be a challenge due to the lack of measurement networks (Yusof, F et al., 2015) (Dhawal Hirani et al., 2016). Predicting seasonal rainfall and its amount of rainfall is a challenging task, and in general there are two approaches to predicting rainfall: they are experiential and operational. The first approach was used to predict the relationship of past historical data, in which recession and artificial neuroscience network approaches use empirical strategy for climate forecasting. To predict seasonal rainfall, physics models and statistical models were used in the second approach. Using ML models such as Regression, Support Vector Machines (SVM) and K-Nearest Neighbours (KNN), researchers have successfully developed predictive and classification models in several fields. However, those models are not successive to attain best accuracy during short-term and long-term rainfall prediction. By considering this goal, the following contributions are presented in this paper.

• Ø From the rainfall dataset, the optimal subset of feature is selected using fuzzy ranking algorithm. It is used to reduce the redundancy feature and computation complexity for rainfall prediction.

• Ø Then, the selected features are taken as input to the optimized DT based rainfall prediction model. In the proposed DT, optimal decision node or upper level of the DT is selected using SOA algorithm.

• Ø Using the proposed SOA based DT prediction model, short-term and long-term rainfall is predicted in the approach.

• Ø The performance of the proposed rainfall prediction model is evaluated in terms of accuracy, F-measure, precision, recall, mean squared error (MSE), Nash–Sutcliffe Efficiency (NSE) and root mean squared error (RMSE).

The following sections are sorted as follows. Rainfall prediction based recent literatures are reviewed in section 2. Section 3 proposes short-term and long-term rainfall prediction using fuzzy ranking algorithm with SOA based DT model. The results of the proposed rainfall forecast are estimated in Section 4 and the conclusion of the paper is described in Section 5.

Background

Venkatesh et al., introduced generative adversarial networks with convolutional neural network for rainfall prediction. In this approach, the authors proposed rainfall prediction using GAN neural network based method. For prediction, LSTM was used as a generator for time series data, and CNN was used for data created to create a discriminative optimal classification. The optimal features were extracted using various feature selection techniques such as Technical Indicators, Fourier Transform, XGBoost, ARIMA, and PCA. From the results, the authors were achieved accuracy of 99% in predicting rainfall.