This dissertation was written as a part of the MSc in Data Science at the International Hellenic University. It studies the effect of the nearest neighbors in timeseries prediction with neural networks. It also provides with an overview of the literature regarding basic deep learning neural networks and timeseries analysis techniques. Most of the concepts described were used in the experimental process.
The goal is to provide insight regarding whether or not the number of nearest neighbors of a timeseries affects the prediction of future values. This gets possible by creating a Long-Short Term Memory model and trying different architectures with the number of nearest neighbors at its center.
During the experimental process, the TISEAN software was widely used to extract val-uable information about the timeseries, such as the minimum and the global embedding dimension of the system, the optimal time delay etc.
Balatsos Christos
07 – 01 – 2022
Collections
Show Collections
