An Application of Markov Chain for Predicting Rainfall Data at West Java using Data Mining Approach

Abstract

Markov chain model is a stochastic process to determine the transition probability of a state space based on a previous state. We can use a stationary distribution of first order-Markov chain model to determine the long terms probability rainfall phenomena. A rainfall data in West Java area has a big data because we can have a large rainfall data from many cities and regencies both of in spatial and time series observations. Furthermore, in this paper, we demonstrate an application of Markov chain using a Data Mining approach to get the knowledge as a pattern for description and prediction the monthly rainfall data in wet seasons December-January-February (DJF) using Knowledge Discovery in Database (KDD) method through pre-processing, data mining process and post-processing. We simulate the monthly rainfall data from the year 1981-2017 using four-state spaces: low (0), medium (1), high (2), and very high (4). The result of Markov chain shows that the probability of occurrence rainfall phenomena for four state spaces are: low (22.62%), medium (24.86%), high (25.46%), and very high is 27.05%. It means that West Java area over the long term condition will have a very high rainfall probability.