Date of Award

Fall 12-7-2023

Degree Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

School

Computing Sciences and Computer Engineering

Committee Chair

Zhaoxian Zhou

Committee Chair School

Computing Sciences and Computer Engineering

Committee Member 2

Chaoyang Zhang

Committee Member 2 School

Computing Sciences and Computer Engineering

Committee Member 3

Bo Li

Committee Member 3 School

Computing Sciences and Computer Engineering

Committee Member 4

Ras B Pandey

Committee Member 4 School

Mathematics and Natural Sciences

Committee Member 5

Sarbagya Ratna Shakya

Abstract

Time series forecasting is a promising technique for various applications which predicts future values or patterns by taking historical data as base. Forecasting future trends is very beneficial for different industries to make valuable decisions and strategies. One such industry is housing market; it has biggest influence on U.S. economy. Housing price index (HPI) is a one of the crucial economic indices published by various government funded and private agency to benefit several industries and individuals for better analysis of future trends of housing market.

Several factors influence the HPI, economical, geographical, and demographic features. Development of traditional time series forecasting models showed greater impact in capturing the trends and seasonality in predicting future values. Advancements in Machine Learning (ML) and Deep Learning (DL) architectures for forecasting using time series data makes the predictions more accurate with a smaller number of data instances and using long term dependencies. Related work shows the prediction of housing market was composed to the local areas based on the geographics and characteristics of the house but not on nationwide economic indicators which makes bigger impact on the market.

In our research, firstly we developed our time series dataset by collecting macroeconomic features which correlates mostly with the HPI from various economic and finance agencies. We also collected the standard dataset of features which predict HPI published by economists every year. Analyzed, visualized, and compared the collected data set with the standard dataset. Designed traditional time series models, machine learning and deep learning models for forecasting HPI using two datasets.

Secondly, compared the performance of the models using evaluation metrics Root Mean Square Error (RMSE), Mean Absolute Error (MAE) and R- Squared (𝑅 2 ). We designed a hybrid architecture by coupling machine learning and deep learning models for better predictions of the targeted variable. Forecasted the future values of HPI using the best performing model

ORCID ID

0009-0008-6471-8421

Available for download on Sunday, December 15, 2024

Share

COinS