Enhanced single exponential smoothing technique for extreme data

Flood is an extreme event that causes damage to properties and loss of human life. Extreme event time series is usually nonlinear pattern. This produces high fluctuations in signal and large uncertainty in forecast quality. Single Exponential Smoothing Technique (SEST) is used for time series forecasting and to smooth out the random variation in signal. However, this technique comes with its own weaknesses to handle extreme data, especially with the presence of outliers. Based on the limitation, the enhancement of the SEST using Membership Function (MF) graph in Fuzzy Logic
(FL) is proposed. The MF graph is used to transform the actual data into new representation which SEST can accept. This technique allows the SEST to use the extreme dataset after the transformation process. The MF is used to transform the actual data into the value in the range of zero to one, while SEST is used to smooth out the irregular component in time series. Six datasets of river water level and three datasets of temperature level were used for this experiment. These techniques have been integrated for handling extreme data. The combination between MF and SEST (MF+SEST) in data smoothing process has shown good performance in handling five river water level data and three temperature data compared to standard SEST. This
combination technique has also improved the performances of Neural Network (NN) model and is capable of producing accurate prediction. MF+SEST has improved the NN performances by an average of 95% compared to SEST and produce more consistent results. This technique can be used in extreme data preprocessing to improve the NN performances to produce an accurate prediction. The prediction
generated will be able to facilitate the authority and public in preparing for upcoming flood event.