Rainfall, Wind Speed, and Temperature Forecast Using Triple Exponential Smoothing and Gradient Descent

The global community strives to minimize the impact of disasters through various actions, for example, mapping flood-prone areas. Flood-prone areas need to be identified correctly, predicted, understood, and socialized to minimize risks when a disaster occurs regarding death, property damage, and socio-economic losses. This type of data-based prediction has been developed and implemented widely and can be applied to predictions related to hydrology. Data mining approaches (estimation, classification, clustering, and time-series forecast) have significantly influenced research related to flood prediction in recent years. The time-series flood forecast has been widely used in previous research using various statistical and data-mining methods. Predicting floods that occur in coastal areas is less discussed than river floods. One method that is often used is exponential smoothing. Determining damping factor values (alpha, beta, and gamma) in the triple exponential smoothing method, in general, is to use all values from 0 to 1 to find the most optimal damping factor, this takes quite a long time and results generally appear with less accuracy. So, a combination of the triple exponential smoothing algorithm is proposed to perform tTimeseries forecast, and the gradient descent algorithm is used as an optimization algorithm to obtain optimal weight values for alpha, beta, and gamma in triple exponential smoothing. Keywords: triple exponential smoothing, gradient descent, flood forecast, flood prediction, time-series forecast