=Paper=
{{Paper
|id=Vol-3910/aics2024_p71
|storemode=property
|title=A Real-Time Prediction System for Restaurant Orders Using Time Series and Behavioural Analytics: A Conceptual Framework
|pdfUrl=https://ceur-ws.org/Vol-3910/aics2024_p71.pdf
|volume=Vol-3910
|authors=Amir Arzy,Fereshteh Khodaparast,Ruairi O'Reilly
}}
==A Real-Time Prediction System for Restaurant Orders Using Time Series and Behavioural Analytics: A Conceptual Framework==
https://ceur-ws.org/Vol-3910/aics2024_p71.pdf
Amir Arzy1,*, Fereshteh Khodaparast2 and Ruairi O’Reilly3
1
Allameh Tabataba'i University, Dehkadeh Olympic Blvd, Tehran, Iran
2
University of Tehran, Enghelab Square, Tehran, Iran
3
Munster Technological University, Rossa Ave, Bishopstown, Cork, Ireland
Abstract
Effective demand forecasting is critical for optimizing operational efficiency within the food-service industry,
which is characterized by inherently volatile consumer preferences and dynamic market conditions. This work
addresses the shortcomings of conventional forecasting methodologies that frequently neglect the integration
of real-time behavioural analytics, thereby leading to inadequate inventory management and diminished
customer satisfaction. The principal aim of this study is to propose a framework for innovative real-time
prediction system that synthesizes advanced time series analysis with behavioural insights garnered from
customer interactions. By employing methodologies such as Seasonal AutoRegressive Integrated Moving
Average (SARIMA) alongside machine learning algorithms, This work seeks to propose a method to improve
demand forecasting accuracy. The proposed framework seeks to amalgamate historical order data with
behavioural analytics, providing restaurant operators with a comprehensive understanding of consumer
demand patterns. This study contributes to the field by introducing a novel approach that not only enhances
forecasting precision but also promotes improved resource allocation and service delivery within the restaurant
sector. This research aims to provide actionable insights to inform strategic decision-making, benefiting
operators and consumers.
Keywords
Real-time prediction, Time series analysis, Behavioural analytics, Machine learning, Demand Forecasting,
Deep Learning
1. Introduction
Catering within the food-service industry is currently facing a pressing need for accurate demand
forecasting. With consumer preferences changing rapidly and market conditions often unpredictable,
maximising operational efficiency while maintaining customer satisfaction is a signi cant challenge.
This work directly addresses the shortcomings of traditional forecasting methodologies, which struggle
to integrate real-time consumer insights and behavioural trends, leading to inefficiencies and potential
revenue losses.
This study addresses inadequate predictive demand forecasting systems within the restaurant sector,
which often need to pay more attention to the dynamic interplay between historical order data and real-
time behavioural analytics. The primary objective is to develop a real-time prediction system that
combines advanced time series (TS) analysis with behavioural analytics derived from customer
interactions and preferences. Unlike existing systems that primarily rely on historical data, the novelty
of the proposed system lies in its ability to incorporate real-time behavioural insights, thereby enhancing
forecasting accuracy and responsiveness. The system is unique in its comprehensive approach, augments
AICS’24: 32nd Irish Conference on Arti cial Intelligence and Cognitive Science, December 09—10, 2024, Dublin, Ireland
*
Corresponding author.
amir_arzy@atu.ac.ir (A. Arzy)
© 2024 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
CEUR
ceur-ws.org
Workshop ISSN 1613-0073
Proceedings
traditional forecasting models and incorporates behavioural data, providing a holistic view of consumer
demand. The proposed approach is intended to lead to optimized resource allocation and improved
service delivery in the food-service industry [1]. The proposed framework focuses on developing a real-
time order prediction model that combines traditional time series methods, such as Autoregressive
Integrated Moving Average (ARIMA) and SARIMA, with Long Short-Term Memory (LSTM) networks
to handle linear and nonlinear patterns in historical order data. Additionally, incorporating behavioural
analytics introduces a new dimension by capturing real-time customer preferences and decision-making
patterns, offering a more dynamic and responsive approach to demand forecasting. By integrating these
elements, the proposed system addresses key limitations in current models. It provides actionable
insights, leading to improved inventory management, targeted marketing, and overall operational
efficiency within the food-service industry.
2. Literature Review
2.1. Advancements in Time Series Analysis for Demand Forecasting Across Industries
The application of TS analysis and artificial intelligence in demand forecasting has gained significant
interest across the healthcare [2], energy, automotive, and food industries, highlighting its contribution
in improving predictive accuracy and boosting operational efficiency. Wang [3] investigates TS
forecasting within the automotive industry, highlighting the integration of K-means clustering with
ARIMA and SARIMA models. This combination enhances forecast precision and emphasizes the
importance of selecting appropriate TS methodologies that align with the unique characteristics of
automotive demand patterns. However, while the study yields promising results, it raises critical
inquiries regarding the model's adaptability to other industries that may exhibit differing demand
dynamics. Matsumoto et al. [4] applies TS analysis to forecast remanufacturing needs in the automotive
parts sector. The work demonstrates TS models capacity to effectively predict demand even in contexts
where complete historical sales data is inaccessible. Nonetheless, the reliance on historical data may pose
limitations, potentially overlooking emerging market trends and shifts in consumer behaviour that are
crucial for accurate forecasting.
In the food service industry, Mihirsen et al. [5] explore the application of machine learning (ML)
systems for forecasting dish sales, emphasizing the role of TS analysis in addressing food waste issues.
Their research suggests that time series-based forecasting models can significantly optimize inventory
management and reduce waste. However, the practical implementation of ML in forecasting hinges on
the quality of historical data and the ability to integrate these systems within existing operational
frameworks, thereby highlighting the inherent challenges of applying advanced TS methods in real-
world scenarios. Mohammed et al. [6] illustrate the potential of employing various TS models for
forecasting clinical laboratory test volumes. While the work showcases the benefits of integrating
statistical techniques in healthcare demand forecasting, it necessitates a comprehensive critique of the
model's applicability across diverse healthcare settings, which can vary due to data availability and
operational constraints.
The exploration of TS methodologies extends to the broader food sector, as evidenced by Suhardi et
al. [7], who examine demand forecasting for roasted coffee production. The approach underscores the
necessity of employing TS techniques that can adapt to shifting consumer preferences and market trends.
However, the study could benefit from a more profound analysis of how macroeconomic factors and
competitive dynamics influence demand, indicating the need for a more comprehensive model.
Concurrently, Kumar et al. [8] applies seasonal ARIMA and exponential smoothing techniques for
electricity demand forecasting, illustrating the effectiveness of seasonal adjustments in enhancing
prediction accuracy while raising questions regarding the robustness of these models in the face of
unpredictable external factors such as regulatory changes and technological advancements in energy
production. Similarly, Panda and Mohanty [9] advocate for integrating regression models with TS
analysis to improve forecasting precision in the food supply chain. Although their findings provide
valuable insights into enhancing forecasting methodologies, the implications of such integrations
warrant further exploration to understand their impact on real-time decision-making processes.
Collectively, these works attest the capacity of sophisticated TS techniques to address the unique
challenges faced in their respective industries. To conclude, the proposed approach recommends a hybrid
model that combines traditional time series algorithms with LSTM networks to enhance the accuracy of
order pattern forecasting in restaurant settings.
2.2. Behavioural Analysis in Demand Forecasting
Demand forecasting has increasingly integrated behavioural analysis to enhance predictive accuracy and
adapt to dynamic consumer preferences. Traditional forecasting models, such as ARIMA and Generalized
Autoregressive Conditional Heteroskedasticity (GARCH), have established a baseline understanding of
TS data; however, the associated limitations in capturing nonlinear relationships and external factors
warrant the adoption of more sophisticated methodologies [10, 11]. Incorporating ML techniques,
particularly hybrid models like Complete Ensemble Empirical Mode Decomposition with Sample
Entropy Support Vector Regression (COEMD-S-SVR), demonstrates significant advancements in demand
forecasting. These models leverage big data and behavioural insights, improving forecasting accuracy
by incorporating real-time inputs from social media interactions and online search patterns [12].
However, the practical implementation of these advanced techniques has its challenges. Managing large
volumes of unstructured data, such as social media posts and online search patterns, can be complex and
time-consuming, complicating the modelling process and hindering effective analysis [13].
Furthermore, ML models such as LSTM networks and ensemble methods have emerged as beneficial
tools for integrating consumer behaviour into forecasting models. LSTMs are particularly adept at
capturing long-term dependencies in TS data, enhancing the predictive power of models when combined
with behavioural signals [14]. Ensemble methods, such as random forests, improve reliability by
aggregating predictions from diverse data sources, thereby mitigating risks of overfitting and enhancing
model performance for short- and long-term forecasts [15]. However, the growing complexity of these
models raises concerns about transparency and interpretability, which are vital for gaining stakeholder
trust, and understanding it, in forecasting outcomes. As decision-makers increasingly rely on these
predictive models, transparency plays a key role in furthering this research.
In addition to behavioural signals, economic and social variables significantly impact demand
forecasting, particularly in volatile sectors such as tourism [16, 17]. The flexibility of statistics models
such as the Grey Model (GM) allows for incorporating external economic variables into time series
forecasting, improving adaptability to changes in consumer behaviour under uncertain market
conditions. Specifically, GM(1,1), a first-order model that predicts a single variable, is mostly used
because it requires limited data while effectively capturing trends [17]. While hybrid models that blend
traditional and ML approaches have shown promise in enhancing forecasting accuracy, achieving a
balance between model complexity and interpretability is an important consideration for practical
applications. Future research should refine these methodologies to capture a broader spectrum of
consumer behaviours and external influences, thus addressing the challenges of increasingly dynamic
market conditions. In summary, integrating behavioural analysis into demand forecasting is a promising
pathway towards more accurate and adaptable predictive models.
Despite advancements in TS analysis and behavioural analytics, a significant gap exists in integrating
real-time data streams for effective demand forecasting. Current models often need a comprehensive
framework that combines these methodologies, particularly in fast-paced environments such as the food-
service industry. Addressing this deficiency is crucial for developing a real-time integrated system
leveraging TS and behavioural analytics.
3. Proposed Methodology
The proposed architecture for integrating behavioural data with TS analysis to enhance predictive
modelling capabilities systematically is depicted in Figure 1. The proposed methodology is explained in
the following stages:
Stage 1: Once the data is collected, it undergoes TS data preprocessing, cleaning and preparing for
analysis to address any inconsistencies or missing values. The workflow progresses to TS modeling and
analysis, where statistical models, followed by TS model evaluation, ensuring the selection of only the
most accurate models for further analysis. In parallel, consumer data is gathered on user interactions
and behaviours through Behavioural Data Collection, which is subsequently preprocessed to ensure it is
structured and ready for analysis.
Stage 2: Following the initial model evaluations, Feature engineering for behavioural data is applied
to derive new features that encapsulate significant behavioural patterns. The Integration of Behavioural
Data with TS merges the processed datasets, resulting in a comprehensive dataset for modelling. Then
this dataset undergoes Model Testing and validation to ensure its robustness and accuracy.
Stage 3: Thereafter, models are ready to be evaluated in a real-world context during the final model
deployment phase. Real-Time data API acquisition facilitates continuous updates and monitoring of the
model's performance. This workflow-based approach ensures a thorough integration of behavioural and
TS data, ultimately leading to meaningful analytical outcomes and actionable insights in real-time
contexts.
3.1. Historical, Behavioural and Real-time Data
Required data must be collected, including historical sales information, order patterns, and inventory
levels. This data will provide insights into customer demand fluctuations over time. Additionally,
contextual information must be gathered such as events, holidays, and weather data, which can
significantly impact ordering behaviour. Obtaining behavioural data from online ordering platforms is
essential for analysing customer behaviour. This data captures critical metrics that provide valuable
insights into consumer preferences and purchasing patterns. Table 1 presents a comprehensive overview
of these key metrics and their descriptions.
Table1: Key Behavioral Data Metrics Captured from Online Ordering Platforms and Their Descriptions
Data Type Description
Order Hesitation Measures the time users take to decide on an order and the
Data frequency of changes before nalizing it—for example,
tracking delays between item selection and checkout.
Session Time This record shows the total duration of a user's session on
the platform, including time spent on product pages, o ers,
and checkout.
Web Page Visits Counts the number of visits to individual menu items or
sections, helping identify popular and less-visited areas on
the platform
Click Data Tracks the frequency and location of clicks throughout the
ordering process, such as clicks on promotions, product
details, or the "Add to Cart" button
Scroll Depth Measures user engagement by recording how far down a
webpage users scroll, indicating interest in speci c sections,
such as item descriptions or reviews
Abandonment Tracks the percentage of users who leave their carts without
Rates completing the purchase, providing insights into potential
barriers in the checkout process
Engagement Evaluates user interactions with promotional content, such
Metrics as banner ads or special deals, by tracking click-through
rates and time spent viewing
Search Behaviour Captures keywords used in search queries, the number of
results viewed, and the corresponding click-through rates,
identifying patterns in product discovery
User Interaction This Record shows actions such as adding items to the cart,
Data modifying selections, navigating between pages, o ering
insights into user preferences
Time of Day Identi es peak activity periods by analyzing user behaviour
Analysis based on the time of day, such as lunch or dinner hours for
restaurant orders
Demographic Data Groups users based on characteristics like age, gender, and
location, providing valuable insights for targeted marketing
strategies
Transactional Data This section details completed orders, including order
frequency, values, and items purchased, o ering insights
into customer loyalty and spending patterns
There are several open-source datasets that can be used as a sample to test the proposed research.
The most useful dataset that researchers found was on the Kaggle website, titled Restaurant Order
Details [18]. This dataset has 500 records for restaurant orders with some customers' demographic
information on a day. However, the limitation of the dataset is that it only records the total number of
items the customer placed, not the details of the menu item they ordered.
After finding an online ordering platform, the system will leverage APIs and web scraping techniques
to continuously monitor and collect relevant behavioural data. This dual approach enables a
comprehensive understanding of customer engagement and preferences alongside the temporal
dynamics of restaurant orders. This ensures a robust dataset that supports effective predictive modelling.
Combining SARIMA and LSTM models offers a robust approach to TS forecasting, particularly
suitable for restaurant order data that often exhibit seasonal patterns and complex nonlinear
relationships. SARIMA effectively captures linear trends and seasonality through its parameters,
represented as SARIMA(𝑝, 𝑞, 𝑑)(𝑃, 𝑄, 𝐷)s, where 𝑝 is the number of autoregressive terms, 𝑞 is the
differencing order, 𝑞 is the number of lagged forecast errors, and the seasonal components are similarly
defined. The SARIMA model can be expressed as follows:
𝜑(𝐵 )(1 − 𝐵 )(1 − 𝐵 𝐵 )𝑌 = 𝜃(𝐵)𝜀
Figure1: Stages of the Proposed Framework for a Hybrid Order Prediction System
𝑌 represents the TS value at time t 𝜑(𝐵 ) and 𝜃(𝐵) are the seasonal autoregressive and moving
average parts, respectively, and 𝜀 is the white noise error term. After fitting the SARIMA model to the
historical data, the residuals are analysed— the differences between observed values and SARIMA
forecasts— to detect any remaining patterns or autocorrelations that the linear model might have
overlooked.
Once the SARIMA model is established its residuals are used as input for the LSTM model. LSTMs
excel at capturing long-term dependencies and nonlinear relationships in sequential data. The
architecture of the LSTM model typically includes input layers, one or more LSTM layers, and output
layers. The residuals, alongside additional features such as promotional events or weather conditions,
are fed into the LSTM model for training. The LSTM can learn to predict these residuals, enabling it to
capture complex behaviours in customer ordering patterns. The final forecasting output is obtained by
summing the SARIMA forecast with the LSTM predictions:
Final Forecast = SARIMA Forecast + LSTM Prediction
This hybrid approach enhances forecasting accuracy by combining the strengths of both models,
addressing the limitations inherent in using each model independently. By effectively modelling linear
trends and complex interactions, the combined SARIMA-LSTM framework can provide more reliable
forecasts, leading to improved decision-making for restaurant operations.
The impact of behavioural analytics on order prediction is profound, as it facilitates a nuanced
understanding of customer interactions and preferences, enhancing forecasting accuracy. Utilizing the
Cross-Industry Standard Process for Data Mining (CRISP-DM) framework, the proposed approach will
address the collection, preprocessing, and analysis of various behavioural data types, including order
hesitation data, session duration, clickstream data, abandonment rates, and demographic information.
These metrics elucidate how customers engage with the online ordering system, highlighting influential
factors in their purchasing decisions. Preprocessing behavioural data is critical to ensuring its quality
and relevance for subsequent analysis. This stage encompasses data cleaning to rectify inconsistencies,
managing missing values, and normalizing features to a standard scale. For instance, the time to finalize
an order may vary considerably across users; thus, normalizing this feature enhances accurate
comparisons across different user sessions. Furthermore, categorical variables, such as user
demographics, will be transformed into numerical formats via one-hot encoding, rendering them suitable
for integration into ML models.
Following preprocessing, feature engineering will concentrate on identifying and constructing
features that exert the most significant influence on the predictive model. This may involve deriving
metrics such as the average order hesitation time over specified intervals, total session duration per
customer, or the frequency of visits to particular menu items. The synthesis of these behavioural features
with TS data enables the modelling of relationships between historical ordering behaviours and future
demand patterns. Researchers can execute this integration using advanced algorithms such as LSTM
networks, which excel in identifying complex patterns within sequential data. By incorporating
behavioural and TS features into the LSTM model, the study aims to capture the interactions between
these domains, thus enhancing the model's capacity to learn from past behaviours and produce accurate
future order predictions. Consequently, this comprehensive approach seeks to improve the precision of
order predictions and aspires to yield valuable insights into customer preferences and behavioural
trends. This empowerment will enable restaurant management to make informed decisions regarding
inventory management, staffing, and targeted marketing strategies, ultimately contributing to
developing robust order prediction systems within the restaurant industry.
In addition to historical and behavioural data, contextual factors such as events, holidays, and
weather conditions must be incorporated into the framework to capture their significant impact on
customer ordering behaviour. For instance, public holidays or major local events often lead to sudden
demand for specific menu items or services. At the same time, adverse weather conditions can shift
customer preferences from dining to delivery options. These fluctuations are critical for businesses to
anticipate, as they directly influence inventory management, staffing, and service efficiency.
Integrating these contextual variables into the predictive model requires advanced preprocessing
techniques to align temporal and spatial data with historical trends and behavioural metrics. This
approach involves handling challenges such as incomplete data, mismatched timeframes, or varying
levels of granularity across datasets.
Additionally, the influence of these factors can vary significantly based on geographical location and
customer demographics. For example, the demand impact of a rainy day may differ in regions
accustomed to frequent rain compared to those where such weather is uncommon. As such, the model
must account for localized and demographic-specific nuances to enhance its adaptability and predictive
power. By effectively incorporating and analyzing these external variables alongside historical and
behavioural data, the framework can better capture dynamic market conditions, thereby improving the
accuracy and relevance of its predictions. This capability is essential for enabling food service businesses
to make informed decisions in real-time, optimizing operations and improving customer satisfaction.
3.2. Limitations of the Proposed Approach
The proposed framework for real-time order prediction in the food service industry presents an
innovative approach by integrating time series analysis with behavioural analytics. However, several
limitations may impact its practical applicability and effectiveness. A key challenge lies in the availability
and quality of data. The framework relies heavily on historical order data and behavioural metrics, but
many food service businesses may need more comprehensive data collection systems, resulting in
incomplete or inconsistent datasets. For example, specific behavioural metrics, such as order hesitation
time or scroll depth, may not be uniformly available across all users or sessions, potentially affecting the
model's reliability.
Integrating behavioural and time series data also presents technical complexities, particularly in
ensuring temporal alignment and consistency between datasets. Discrepancies in timestamps,
differences in data granularity, and the need for extensive preprocessing increase the computational
effort required. Furthermore, the hybrid approach, which combines statistical models like SARIMA with
deep learning models such as LSTMs, demands substantial computational resources. Training and testing
these models on large-scale datasets can be resource-intensive, which may pose a barrier for small or
resource-constrained businesses.
Another limitation arises from the dynamic nature of external factors. While the framework
incorporates behavioural data, it does not explicitly account for variables such as sudden market changes,
promotional events, or macroeconomic trends that could significantly impact predictions. This issue
could result in reduced accuracy in highly volatile scenarios. The framework's scalability and
generalization also pose challenges, as it is specifically designed for the food service industry. Expanding
its application to larger enterprises or other sectors with unique behavioural patterns may require
significant customization and retraining.
Finally, the model's interpretability is a critical consideration. Advanced machine learning techniques
and intense learning models like LSTMs often function as "black-box" systems, making it difficult for
stakeholders to understand the rationale behind predictions. Improving model interpretability will be
essential for building trust and facilitating adoption among decision-makers. Addressing these
limitations through future research will enhance the framework's practicality, scalability, and
effectiveness in real-world applications.
4. Results and Discussion
Integrating behavioural analytics with TS modelling presents a transformative approach to demand
forecasting that offers considerable advantages over traditional systems relying exclusively on TS data.
Established models such as SARIMA and LSTM excel at capturing historical trends, seasonal variations,
and external factors like holidays and weather conditions. However, these models often need to be
revised in their adaptability to abrupt shifts in consumer behaviour. For instance, during peak dining
seasons or local events, customer preferences can shift rapidly, resulting in demand spikes or drops
requiring adequate historical data reflection. Consequently, traditional TS models may produce
inaccurate predictions, resulting in inefficient resource allocation, staffing issues, and potential loss of
customer satisfaction due to longer wait times or out-of-stock items.
In contrast, the proposed hybrid model enhances demand forecasting by incorporating behavioural
data, enabling a more dynamic and responsive forecasting mechanism. The proposed model captures
real-time insights into customer interactions and preferences by utilizing behavioural data such as order
hesitation data, session duration, clickstream analysis, and engagement metrics. This additional layer of
information allows for improved predictions of demand fluctuations, as it considers historical sales data
and immediate factors influencing consumer decision-making. The proposed model enables restaurants
to forecast demand, allowing them to place orders for ingredients or prepare dishes based on predicted
customer behaviour before receiving actual orders. For example, the proposed model can identify
patterns in user engagement that precede increases in orders, such as a rise in menu item views or
heightened interaction with promotional offers.
The distinctive nature of the proposed approach lies in its combination of behavioural analytics with
TS forecasting, which sets it apart from existing methodologies that primarily focus on historical sales
data alone. While many traditional models rely solely on past performance metrics, this study
emphasizes the importance of real-time behavioural indicators in enhancing demand predictions. By
incorporating these elements, the proposed model offers insights into the immediate drivers of consumer
behaviour, allowing for proactive adjustments in restaurant operations. Moreover, these improvements
are aligned with the European Union's sustainability goals, particularly to reduce food waste and end
hunger by 2030. The proposed framework will offer a significant advancement in demand forecasting,
providing restaurants with the tools to optimize resource allocation, order preemptively based on
predicted demand, and improve customers' overall dining experience - thereby contributing to
operational efficiencies and broader sustainability goals.
5. Conclusion and Future Work
This proposed framework introduces an innovative approach to real-time order prediction in the food
service industry by integrating time series analysis with behavioural analytics. The model addresses
significant gaps in current predictive models by combining dynamic time series forecasting with
customer behavioural insights. Through this hybrid approach, the model aims to enhance the accuracy
and adaptability of demand forecasting, offering real-time solutions to operational challenges such as
inventory management and order fulfilment.
Despite advancements in predictive modelling across various sectors, including retail, healthcare, and
energy, hybrid models are still limited in their application to real-time order prediction in the food
service industry. The full potential of behavioural and historical order data in improving forecasting
accuracy has yet to be fully realized. This work seeks to bridge this gap by integrating behavioural and
historical order data to improve prediction precision.
Future work will focus on validating the proposed model in real-world food service settings. This
approach includes collecting and integrating real-time behavioural data, followed by pilot studies to
assess the model's effectiveness in predicting order volumes. To enhance the robustness of the model,
further exploration of advanced machine learning techniques and the inclusion of additional external
factors, such as market trends and competitor analysis, will be pursued. The framework's scalability
across different food service environments will also be evaluated to refine the model for broader industry
applications and further operational optimization.
References
[1] L. Xiao, J. Wu, and Z. Zhang. Integrating Behavioural Analytics and Time Series Forecasting for
Restaurant Demand Prediction: A Big Data Approach. *Journal of Business Research*, 128: 511–520,
2021.
[2] P. Nisha, U. Pawar, R. O'Reilly, Interpretable machine learning models for assisting clinicians in the
analysis of physiological data, in: Proceedings of the AICS Conference, 2019, pp. 434–445.
[3] Y. Wang. Time Series Forecasting in the Automotive Industry: A Case Study. *Automotive Science
and Technology*, 27(2): 215–230, 2023.
[4] S. Matsumoto, M. J. Kim, and T. Fujimoto. Time Series Demand Forecasting for Remanufacturing.
*International Journal of Production Economics*, 170: 303–313, 2015.
[5] S. Mihirsen, A. Kumar, and N. Sharma. Machine Learning Applications for Food Waste Reduction: A
Case Study. *Waste Management*, 103: 136–144, 2020.
[6] E. A. Mohammed, C. Naugler. Open-source Software for Demand Forecasting of Clinical Laboratory
Test Volumes Using Time-series Analysis. Journal of Pathology Informatics, 2017.
[7] H. Suhardi, H. Suharno, and H. Pranoto. Forecasting Demand for Roasted Coffee Production Using
Time Series Analysis. *Journal of Agricultural and Resource Economics*, 44(1): 77–89, 2019.
[8] P. Kumar, and R. Singh. Restaurant Demand Forecasting: A Systematic Review and Future Research
Directions. *Journal of Hospitality and Tourism Management*, 39: 49–59, 2019.
[9] K. Panda, and S. P. Mohanty. Integrating Regression Models with Time Series Analysis for Food
Supply Chain Forecasting. *Food Quality and Preference*, 92: 104–112, 2023.
[10] J. Fan, and Y. Hong. Time Series Analysis: A Comprehensive Review. *Journal of Time Series
Analysis*, 43(2): 215–234, 2022.
[11] O. Claveria, and S. Torra. Forecasting Tourism Demand with Linear and Nonlinear Models: A
Comparative Study. *Tourism Management*, 42: 214–226, 2014.
[12] S. Seyedan, A. J. Molla, and M. Sadeghian. COEMD-S-SVR: A Hybrid Model for Demand
Forecasting Based on Social Media. *Journal of Business Research*, 120: 451–464, 2020.
[13] A. Arzy, M. T. Taghavifard, Z. Dehdashti Shahrokh, I. Raeesi Vanani, An investigation on the
user behavior in social commerce platforms: a text analytics approach, Journal of Information
Technology Management 12(4) (2020) 180–199.
[14] L. Hong, Y. Wang, and Z. Wang. Long Short-Term Memory Network for Time Series Prediction:
A Review. *Applied Sciences*, 11(8): 3632, 2021.
[15] G. Assaf, C. P. Barros, and C. H. Hsu. Machine Learning Techniques for Demand Forecasting:
An Overview. *International Journal of Hospitality Management*, 81: 30–39, 2019.
[16] H. Song, and G. Li. Tourism Demand Forecasting: A Review of Time Series Models. *Tourism
Management*, 29(2): 319–338, 2008.
[17] X. Liu, J. Zhang, and J. Chen. Economic Variables and Consumer Behaviour in Time Series
Forecasting. *Journal of Economic Dynamics and Control*, 136: 104–117, 2023.
[18] Harris, M. (2018). Restaurant Order Details [Data set]. Kaggle.
Retrieved from https://www.kaggle.com/datasets/mohamedharris/restaurant-order-details