Digital Prophet: Predictive Analytics of Time Series Datasets Using Hyperparameter Optimization
Project Author
Issue Date
2023-04-28
Authors
Coleman, Kobe
Loading...
Embargo
First Reader
Jumadinova, Janyl A.
Additional Readers
Kapfhammer, Gregory
Keywords
Machine Learning , Neural Networks , Hyperparameters , Long Short-Term Memory
Distribution
Abstract
This project is designed to optimize the configurations of neural networks. Neural networks are widely used in various industries due to their ability to identify complex patterns in data and produce accurate output, including predictions. They are particularly useful in scenarios where it is difficult for humans to accurately analyze and predict outcomes based on the available data. The configurations of a neural network are called hyperparameters, and they play a large role in neural network performance. As it stands, finding the hyperparameter values for the most optimal neural network is a challenging process, as it involves trial and error. This requires anyone who constructs neural networks to deal with tradeoffs between performance and optimality. This project focuses on addressing this issue. The artifact in the project takes a dataset as input, creates a container of models with
varying hyperparameter values, and provides as output the model that had the best performance. The experiment in the project runs the artifact multiple times and creates a series of findings organized in multiple plots that can be referred to when constructing neural networks that are designed to create forecasts from time series data. By using those findings, the amount of trial and error typically required to construct the most optimal neural network while still maximizing performance can be significantly reduced.
Description
Collections
Chair
Major
Computer Science
Department
Computer Science
Recorder
License
Citation
Version
Honors
Computer Science, 2023