A novel approach of hidden Markov model for time series forecasting

Abstract

In the past few years, tremendous studies have been made to examine the accuracy of time series forecasting that provide the foundation for decision models in foreign exchange data. This study proposes a novel approach of Hidden Markov Model and Case Based reasoning for time series forecasting. This paper compares the proposed method with the single HMM and HMM ensemble with neural network. HMM is trained by using forward-backward or Baum-Welch algorithm and the likelihood value is used to predict future exchange rate price. The forecasting accuracy has been measured according to Root Mean Square Error (RMSE). The statistical performance of all techniques is investigated in testing of EUR/USD exchange rate time series over the period of October 2010 to March 2014. The preliminary results indicate that the new approach of HMM produce the lowest RMSE compared to the benchmark models. Further study is to adopt HMM-CBR in testing of GBP/USD, GBP/JPY, USD/JPY, and EUR/JPY exchange rate. KeywordsHidden Markov Model, Case Based Reasoning, Baum-Welch algorithm, Ibk algorithm. INTROUCTIONForecasting time series has been one of the most challenging tasks due to non-linear in nature. Hidden Markov Model (HMM) plays a key role to find the non-linear patterns [1].Over the past few decades, HMM has been widely applied as a data-driven modeling approach in automatic speech recognition [2].HMMs applications are now being extended to many fields such as traffic congestion [3], dynamic system modeling and diagnosis [4], neurosciences [5], stock market [1,6,7,8,9,10,11,12], inflation analysis [13], safety messages [14] and exchange rate [15, 16]and etc. Hidden Markov Model basically is a statistical model that has been proven useful in wide range applications domain. On the other hand their several limitations and mix results regarding their forecasting rule are creating vagueness about their use in real exchange rate trading environment. This scenario is promoted by the fact that the initialization of model parameters and inputs is difficult because of many factors that leads to diverse conclusions. According to [16] HMM are unstable to be taken in as a trading tool on foreign exchange data with too many factors influencing the results [15]. Diana Roman and Gautam Mitra in [20] consider estimating the parameters of an HMM is a difficult task. While implementing the HMM, the choice of the model, choice of the number of states and observation symbol (continuous or discrete or multi-mixture) become a tedious task [6]. Continuous Hidden Markov Model has high calculation complexity problem [10].This paper is to establish a new approach of Hidden Markov Model which attempt to overcome some of the limitations from the literature. In particular the proposed method is to reduce the numbers of parameters while on the other hand it increases the forecasting ability of the model. The study used the well established technique of HMM in a new way to forecast EUR/USD currency pair. In this paper, the first step is to establish pattern from the p...

User profiling using smartphone network traffic analysis

“…Hidden Markov models present an approach modeling future states of a stochastic system based on observed or hidden state variables. This sets the framework for predictive sequential algorithms which have been proven robust [8,9,10] at modeling features like prices, volatility and regime classification. Fu et al [11] describe a framework that labels traffic based on usage type (such as text, multimedia, location sharing) using an ensemble classifier model where a base classifier proposed as a random forest is boosted with outlier detection using combined k -means and Hidden Markov models.…”