EMERGING STATISTICAL MACHINE LEARNING TECHNIQUES FOR EXTREME TEMPERATURE FORECASTING IN U.S. CITIES - A PREPRINT

Aug 5, 2026

Upload by Patrick

The paper analyzes extreme temperature patterns through emerging statistical machine learning techniques, aiming to compare model effectiveness for climate time-series forecasting. It evaluates ARIMA, exponential smoothing, multilayer perceptrons, and Gaussian processes on data from five of the most populated U.S. cities using Python and Julia. Results show clear differences among methods and identify multilayer perceptrons as the strongest option. Using this model, the study forecasts extreme temperatures through 2030 and tests whether projected temperature changes exceed zero.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages13

Reading Time33 min

AccessFree to download

Introduction

Statistical models overview

Statistical Models

Auto-Regressive Integrated Moving Average

Exponential Smoothing

Multilayer Perceptron

Gaussian Processes

EMERGING STATISTICAL MACHINE LEARNING TECHNIQUES FOR EXTREME TEMPERATURE FORECASTING IN U. S. CITIES A PREPRINT arXiv :2307 . 14285v1 [ stat .AP] 26 Jul 2023 Kameron B. Kinast School of Mathematics and Statistics Rochester Institute of Technology Rochester, New York 14623 [kbk7499@rit.edu](kbk7499@rit.edu) Ernest Fokoué School of Mathematics and Statistics Rochester Institute of Technology Rochester, New York 14623 [epfeqa@rit.edu](epfeqa@rit.edu) July 27, 2023 ABSTRACT In this paper, we present a comprehensive analysis of extreme temperature patterns using emerging statistical machine learning techniques. Our research focuses on exploring and comparing the effectiveness of various statistical models for climate time series forecasting. The models considered include Auto-Regressive Integrated Moving Average, Exponential Smoothing, Multilayer Perceptrons, and Gaussian Processes. We apply these methods to climate time series data from ﬁve most populated U.S. cities, utilizing Python and Julia to demonstrate the role of statistical computing in understanding climate change and its impacts. Our ﬁndings highlight the differences between the statistical methods and identify Multilayer Perceptrons as the most effective approach. Additionally, we project extreme temperatures using this best-performing method, up to 2030, and examine whether the temperature changes are greater than zero, thereby testing a hypothesis. Keywords time series 􀀁 weather forecasting 􀀁 statistical models 􀀁 climate science 1 Introduction For many decades, climate scientists have been concerned about the impact of climate change. Climate change increases the risk to both natural and human systems, and the degree of risk depends on various factors, including extreme hot temperatures [1] . The 2015 Paris Agreement was established to limit the increase in temperatures to 1.5􀀎 C above pre-industrial levels (1850-1900) [2] . Achieving this goal would reduce exposure to climate-related risks, such as heatwaves, droughts, and extreme precipitation events [1] . Understanding, modeling, and forecasting weather patterns pose ongoing challenges. Time series forecasting is a technique used to predict future observations by analyzing past values. Scientists employ various machine learning and traditional approaches to analyze and predict these events. For example, Kumar and Middey [3] used a hybrid of random forest and autoregressive integrated moving average (ARIMA) model to project extreme climate indicators. The objective of this paper is to provide a comprehensive review of the effectiveness of commonly used classical and machine learning methods for time series forecasting. Numerous comparative studies have compared classical and machine learning methods for time series forecasting. For instance, Hill and other colleagues compared the neural network model with classical models such as the Box-Jenkins model, single exponential smoothing model, and naive model, and claimed that the neural network model performed the best[4] . Nesreen and other authors [5] conducted a comparative study of different machine learning models for time series forecasting and found that multilayer perceptron (MLP) and Gaussian processes (GP) regression were the best models for forecasting with M3 competition data or different types of time series data. Since classical methods still hold signiﬁcance in time series analysis and forecasting, this paper will study four different time series forecasting methods: 1. Auto-Regressive Integrated Moving Average 2. Exponential Smoothing 3. Multilayer Perceptron 4. Gaussian Processes As mentioned, the Paris Agreement aims to limit the temperature increase to 1.5 degrees Celsius to mitigate the vulnerability to severe climate effects. This research paper investigates the rate of temperature change over a 28-year period (2002-2030) . Hypothetically, this paper performs time series forecasting to test the hypothesis that

FAQs

Which forecasting methods are compared for extreme temperature prediction?

The paper compares ARIMA, exponential smoothing, multilayer perceptrons, and Gaussian process models for climate time-series forecasting.

What dataset and cities are used in the study?

The methods are applied to climate time-series data from five of the most populated U.S. cities.

How far into the future does the paper forecast extreme temperatures, and what hypothesis is tested?

Extreme temperatures are projected up to 2030 using the best-performing model, and the paper tests whether the temperature change rate is greater than zero.

Aug 5, 2026

Upload by Patrick