Comparing Forecasting Models for Univariate Time-Series Analysis

Modern deep learning architectures have become powerful tools for prediction and time-series forecasting, being able to reach high accuracy metrics for high dimensional, large sample datasets. However, they are not without their limitations, as they are particularly prone to overfitting for relatively smaller sample sizes. In this case, we turn to traditional time-series forecasting methods, and find that they perform better than deep learning architectures for “small” sample size forecasting.

This work was done under the UW course CSE 599G1: Deep Learning for Computer Vision, taught in Spring 2026.

The written report can be found here, with it’s corresponding poster.