Forecasting is a broadly applicable field, from financial gains, over energy, weather, or political votes. Everyone seems to want to know the future. In this thesis project, we want to have a look at machine learning forecasting techniques from an interdisciplinary perspective. In machine learning based forecasting, we train a network on historical data to predict future data.
Maybe you have head of approaches such as ARIMA, Recurrent Neural Networks, LSTMs, or Transformers. But which one is really the best?
Forecasting tasks are not only characterized by the data domain, but also by what we want to know about the future: Do you want to predict the next hour, the next day, or the next year? Depending on the goal, task complexity may differ widely and so may the best architecture.
In this thesis project, you will compare different machine learning architectures for forecasting tasks. Due to its broad intent, you will have the opportunity to develop insights into a wide range of forecasting tasks and their common and unique challenges.
Maybe you have head of approaches such as ARIMA, Recurrent Neural Networks, LSTMs, or Transformers. But which one is really the best?
Forecasting tasks are not only characterized by the data domain, but also by what we want to know about the future: Do you want to predict the next hour, the next day, or the next year? Depending on the goal, task complexity may differ widely and so may the best architecture.
In this thesis project, you will compare different machine learning architectures for forecasting tasks. Due to its broad intent, you will have the opportunity to develop insights into a wide range of forecasting tasks and their common and unique challenges.