Advancing Time Series Forecasting: A Comparative Study of Mamba, GRU, KAN, GNN, and ARMA Models

Author(s): Shenggang Li

Originally published on Towards AI.

Evaluating Modern and Traditional Methods for Multivariate Time Series Prediction

This member-only story is on us. Upgrade to access all of Medium.

Imagine trying to predict how a group of related factors change over time — like stock prices that affect each other, gene activity in a biological system, or sales of connected products in a supply chain. These aren’t just separate time series; they interact and influence one another in ways traditional models often miss. This is what makes multivariate time series forecasting both fascinating and challenging.

For years, models like ARMA have been the standard for time series forecasting. They work well for handling one series at a time but struggle to capture relationships between multiple series. In reality, that’s a big gap. Stock prices are connected, economic indicators move together, and the sales of one product can impact the entire inventory. What we need are models that not only track individual trends but also understand how these series influence each other.

In this paper, I explore five different approaches for tackling multivariate forecasting:

The Mamba-inspired model for capturing time-based dynamics.GRU (Gated Recurrent Units) for finding patterns in sequences.Kolmogorov-Arnold Networks (KAN) with dynamic weights to uncover nonlinear relationships.Graph Neural Networks (GNNs) to map relationships between series as… Read the full blog for free on Medium.

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI