Multi-Level Hidden Markov Model and USLTM Network

Abstract

Financial data is a representative example of time series data. In analyzing time series data, unlike other data types, observations at other points of times act primarily to interpret the current observations.Time series data have been studied for a long time using traditional methodologies. This thesis present methods analyzing time series data especially the financial data. Several experiments presented in this thesis will show the effectiveness of the introduced machine learning models. This thesis cover not only a classical machine learning techniques but also recently active techniques. \\

Time series data is one of important subjects of machine learning. Compared to classical methods, Machine Learning has had an remarkable effect in analyzing time series. We will describe some time series analysis methods that are typical for machine learning. We will also present more advanced models. The first one of them is a model that uses the Markov chain. Chapter 2 provide the basic knowledges about the Markov chains. In Chapter 3, we present an existing model whose base is on the Markov chains. In consequent chapter, a new model that we created will be introduced. The experimental results are also contained in the chapter.\\

The second part of this thesis start from explaining the deep learning architecture. Chapter 5 contains explanations about basic notations in deep learning and specific type of models in deep learning architecture. The models introduced in this chapter are often used when dealing with time series data in deep learning. In Chapter 6 we present an extended version of the model based on the models introduced in Chapter 5. In this chapter, we conduct an experiment to compare our model with the existing model.