Mathematical Engineering Design Thesis Project

Project Name

• Deep Learning Algorithms for Time Series Analysis with Applications

Team Members

• Eda Atalay
• İpek Korkmaz

Introduction

In this project, feed-forward neural networks and recurrent neural networks, which are a subject of deep learning was thoroughly examined. For a better understanding of deep learning, literature research has been done and the history of deep learning was discussed. In the project, especially the mathematical relationships and the related equations were presented. Also, a special type of recurrent neural network for time series, LSTM was explored in depth.

In application, the data set containing 44 county pairs, export values for each country pair, and several factors that may affect export values, was examined. The dataset and the meaning of those factors were thoroughly discussed.

The main purpose of the project is to investigate export values change over time and to predict the future. For that reason, several models have been built on Python. Three models have been built using LSTM and the other three models have been built with feed-forward neural network. In order to determine the optimal model for data set, model error values have been compared and regarding this comparison, the discussion has been made.

Instructions & Explanations

• You can display the project's overview and detailed information about mathematical relationships behind neural networks and LSTM by viewing deployment on GitHub Pages.
• You can display the model building part, codes and information about dataset by viewing "application_model_building.ipynb" file.