TLDR: A study utilized Long Short-Term Memory (LSTM) networks to forecast hotel Key Performance Indicators (Occupancy, Average Daily Rate, Revenue per Available Room) across Manchester, Amsterdam, Dubai, Bangkok, and Mumbai using monthly data from 2018-2025. The research found that LSTM models provide highly accurate forecasts, with Manchester and Mumbai showing the most predictable patterns due to stable demand, while Dubai and Bangkok exhibited higher variability due to seasonal and event-driven influences. The findings validate LSTM’s effectiveness for urban hospitality forecasting and offer a comparative framework for data-driven decision-making for tourism stakeholders and urban planners.
The global urban hospitality industry is constantly changing, influenced by factors like rapid urbanization, evolving traveler expectations, and major events such as the COVID-19 pandemic. In this dynamic environment, accurately predicting hotel performance is crucial for managing revenue, planning strategically, and making informed policy decisions. Key Performance Indicators (KPIs) like Occupancy (OCC), Average Daily Rate (ADR), and Revenue per Available Room (RevPAR) are essential metrics used by hoteliers and stakeholders to evaluate market performance.
Traditionally, forecasting in the hospitality sector relied on classical time series techniques such as ARIMA and exponential smoothing. While these methods work well in stable conditions, they often struggle to capture the complex, non-linear, and seasonal patterns found in modern urban hotel markets. Recent advancements in computing power and data availability have shifted the focus towards machine learning and deep learning approaches for hospitality forecasting.
Among these advanced techniques, Long Short-Term Memory (LSTM) networks, a type of Recurrent Neural Network (RNN), have demonstrated superior performance in time series forecasting. LSTMs are particularly effective at handling long-term dependencies and non-linear trends, making them well-suited for the complexities of the hotel industry where temporal dependencies, seasonality, and sudden changes often coexist.
A recent study, titled “Deep Learning-Based Forecasting of Hotel KPIs: A Cross-City Analysis of Global Urban Markets”, addresses a significant gap in existing research by conducting a cross-city analysis of hotel performance forecasting using LSTM models. This research focuses on five major global urban markets: Manchester, Amsterdam, Dubai, Bangkok, and Mumbai. These cities were chosen for their diverse geographic locations and varying tourism dynamics, ranging from established Western destinations to rapidly growing Asian cities. For a deeper dive into the methodology and detailed findings, you can read the full research paper here.
Study Methodology
The study utilized monthly data from 2018 to January 2025, focusing on core hotel performance indicators from STR Global: Occupancy Rate (%), Revenue per Available Room (RevPAR), and Average Daily Rate (ADR). Data preprocessing involved handling missing values, outliers, and applying normalization techniques to ensure consistency. The data was then structured into monthly time series formats.
The forecasting model was built using LSTM networks, which are designed to manage sequential data with long-range dependencies. Unlike traditional RNNs, LSTMs use a memory cell structure with input, output, and forget gates, allowing them to effectively retain or discard information over time. Hyperparameters for the LSTM models were fine-tuned using a combination of grid search and Bayesian optimization to enhance performance.
Two forecasting strategies were employed: rolling forecasting to assess the model’s adaptability over time, and multi-step forecasting to project future KPI values at 3-, 6-, and 12-month intervals. Model performance was evaluated using Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE). The data was split, with 80% (from 2018 to late 2023) used for training and the remaining 20% (late 2023 to 2025) for testing, ensuring a robust evaluation of the model’s out-of-sample performance. This cross-city validation confirmed the model’s adaptability and potential for wider application in global urban hospitality analytics.
Key Findings and Performance Insights
The study revealed distinct pandemic responses and recovery patterns across the five cities. Pre-pandemic, Dubai and Amsterdam led in occupancy rates, while Bangkok and Dubai were hit hardest during the pandemic due to their tourism dependence. Dubai, however, showed remarkable resilience, surpassing pre-pandemic occupancy levels by 2023, driven by luxury tourism and events like Expo 2020. Mumbai also recovered strongly, benefiting from returning business travel. Manchester maintained stability through domestic demand, while Amsterdam lagged with slower business travel returns, and Bangkok struggled to exceed 75% occupancy even by 2024.
In terms of Average Daily Rate (ADR), Amsterdam, initially the highest-priced market, suffered the most severe pandemic shock and its recovery remained incomplete by 2025. Dubai, despite a significant drop, not only fully recovered but surpassed pre-COVID rates. Bangkok, starting as the lowest-priced market, experienced a dramatic collapse but mounted an impressive recovery. Manchester showed remarkable stability with the shallowest decline, while Mumbai emerged as a standout performer, reflecting India’s economic ascent.
For Revenue per Available Room (RevPAR), Amsterdam experienced the steepest decline and showed lasting challenges in regaining its premium position. Dubai demonstrated remarkable resilience, achieving full recovery and surpassing pre-pandemic levels. Bangkok, after a complete collapse, mounted a strong recovery. Manchester showed the steadiest performance, exceeding pre-pandemic RevPAR by 2023 due to its domestically driven market. Mumbai emerged as a growth champion, surging significantly by 2025, underscoring India’s economic momentum.
The LSTM model’s forecasting performance was evaluated using Mean Squared Error (MSE) and Mean Absolute Percentage Error (MAPE). According to widely accepted categorization, MAPE values below 10% indicate highly accurate forecasting. The models developed in this study consistently exhibited strong predictive performance across all cities and KPIs, with most MAPE values well below 10%.
Manchester consistently showed low MSE and MAPE across all KPIs, indicating high accuracy in forecasting its hotel performance, likely due to stable demand patterns. Mumbai demonstrated the lowest MAPE for ADR, suggesting stable and predictable price trends in its hotel sector. Conversely, Dubai showed the highest MSE and MAPE for RevPAR, implying greater complexity or unpredictability in its revenue patterns, possibly due to fluctuations driven by seasonality, high-end tourism, and event-based demand. Bangkok’s ADR forecasting also presented challenges, reflecting variability in pricing strategies or external economic influences.
Also Read:
- AI-Powered Energy Management for Healthcare Facilities: A Case Study in Demand Forecasting and Load Balancing
- REP-Net: A Modular Framework for Advanced Time Series Forecasting
Conclusion
This study effectively demonstrates the power of advanced machine learning techniques, specifically LSTM networks, in forecasting hotel performance across diverse urban environments. By analyzing Occupancy, Average Daily Rate, and Revenue per Available Room for five globally significant cities, the research highlights both the strengths and challenges of forecasting in different market contexts. The findings confirm that LSTM models, when supported by robust data strategies, offer a reliable and scalable approach for urban hospitality forecasting. These insights are invaluable for hoteliers, policymakers, and tourism strategists, aiding them in operational planning, investment decisions, and post-crisis recovery efforts. The model’s adaptability across various urban profiles underscores its broad relevance for data-driven hospitality management worldwide.
