Keywords

Artificial Intelligence; Machine Learning; Deep Learning; Neural Networks; Transformers; Binary Classification

Abstract

Airbnb hosts and renters lack reliable tools to anticipate future occupancy, which is critical for setting competitive prices, planning availability, and making informed booking decisions. This is especially true as artificial intelligence capabilities continue to advance, so focusing on using the most efficient and effective models is vital. This is especially important when considering an average user, who might want a specialized model but has limited resources. By using the correct models, accuracy and computing time can be optimized for the end user. This thesis addresses that practical gap by evaluating several state-of-the-art machine learning models for predicting Airbnb listing occupancy, to produce recommendations that balance prediction quality and practical computing efficiency. We use the random forest classifier, a simple two-layer neural network, a time series transformer, BERT, TCN, and xLSTM as machine learning models. We begin by fine-tuning each model using hyperparameter sweeps to ensure fair comparison through proper optimization. We then use different cities’ Airbnb data to explore the effects of the dataset’s size and temporal density on each model’s results. Understanding these effects is crucial for determining which models perform best under specific conditions. Lastly, we identify limitations of our experimentation and outline future challenges that can be addressed in this field.

Thesis Completion Year

2025

Thesis Completion Semester

Fall

Thesis Chair

Dutta, Aritra

College

College of Engineering and Computer Science

Department

Department of Computer Science

Thesis Discipline

Computer Science

Language

English

Access Status

Open Access

Length of Campus Access

None

Campus Location

Orlando (Main) Campus

Download

Supplemental Files

Share

COinS
 

Rights Statement

In Copyright