Publication Date

Summer 7-29-2025

Abstract

The increasing rate of drug overdose deaths in the United States poses a critical public health challenge, particularly due to the surge in synthetic opioids and other high-risk substances. This study presents a data-driven framework that integrates time series forecasting and clustering techniques. Monthly mortality data for five key drug types: cocaine, fentanyl, heroin, methamphetamine, and oxycodone were analyzed using four time series forecasting models: ARIMA, ETS, TBATS, and NNAR. These models were evaluated using standard accuracy metrics RMSE, MAPE, and MAE to assess predictive performance. Signal decomposition approach based on Singular Value Decomposition and subspace modeling was employed to decompose the Fentanyl series, revealing the dominant components and underlying trends. Clustering analyses were conducted to identify regional similarities in drug overdose mortality patterns. K-Means and hierarchical clustering methods were applied to group regions with similar trajectories, and a co-clustering frequency matrix assessed the consistency of these groupings across drug types. This integrated approach enhances understanding of both temporal and spatial patterns in drug-related mortality, providing valuable insights for designing targeted public health interventions.

Degree Name

Mathematics

Level of Degree

Masters

Department Name

Mathematics & Statistics

First Committee Member (Chair)

Stephen Lau

Second Committee Member

James Degnan

Third Committee Member

Miheer Dewaskar

Language

English

Keywords

Time Series, ARIMA, ETS, TBATS, NNAR, Clustering, K-Means, Decomposition

Document Type

Thesis

Download

Share

COinS