| Summary: | 碩士 === 國立臺灣科技大學 === 工業管理系 === 107 === With the advancement of today's technology, the speed of data generation is getting faster and the amount of data is getting larger. The concepts of Big Data have progressed from 3Vs to 5Vs. Nowadays, Artificial Intelligence, Machine Learning and Cloud Computing play important roles in data science. In this thesis, we focused on exchange rate prediction from financial Big Data with TensorFlow and Keras, and used Minitab to analyze the results.
In this study, we divided the research process into two phases. The first phase was data prediction. First, data preprocessing converted the collected exchange rate data into a usable format and normalization. After preprocessing, we divided the data into training, validation and testing data sets. Next step, we used TensorFlow as the back-end engine in Keras to construct a one-dimensional Convolutional Neural Network (CNN) model with a sliding window for predicting exchange rate. We selected the model parameters by the design of experiment method. After choosing the best parameters for the CNN model, we used the testing data set to do the performance evaluation. Finally, we forecasted the exchange rate for 2 years. The second phase was the result analysis. We put historical data combined with the prediction values into Minitab software and used the two-sample t test to check whether there was a significant difference between the previous year and the next year. After analysis, we could exam which period and which country control the exchange rate. Furthermore, this modeling process was not limited to this case data, and more influencing factors of the exchange rate could be added in this study to achieve analytical integrity.
|
|---|
