Dynamic Price Prediction for Event Services Using Machine Learning: A Vendor-Centric Approach

Dynamic Price Prediction for Event Services Using Machine Learning: A Vendor-Centric Approach

Priya H

Department of MCA

St. Joseph Engineering College, Mangalore, India

Nishmitha J

Department of MCA

St. Joseph Engineering College, Mangalore, India

Sumangala N

Department of MCA

St. Joseph Engineering College, Mangalore, India

Jessen Avin Department of Computer Applications

St. Joseph Engineering College, Mangalore, India

Kaushik B Department of MCA St. Joseph Engineering

College, Mangalore, India

Murari B K Department of Computer Applications

St. Joseph Engineering College, Mangalore, India

AbstractPricing in the event industry can be tricky, as it is influenced by factors like seasonal demand, event type, and the vendors experience level. Traditional fixed pricing often misses opportunities to optimize profits or adjust to real-time market needs. Traditionally, static pricing can create revenue opportunities, and pricing efficiencies are often missed. In this work, we illustrate a machine learning-based price prediction system that utilizes raw vendor data to produce optimal price recommendations that are adaptive. The prediction model in- cludes regression-based models including Linear Regression, Decision Tree and XGBoost to estimate service costs based on event category, location, seasonality, and vendor experience. Performance assessment of the prediction model relies on RMSE and MAE. The recommendation system is developed in Python and the MongoDB is utilized for data storage. The experimental results of the price prediction system indicate better price consistency and overall vendor profit enhance – a shift in thinking from static pricing appears warranted and data directing price improvements in the event management sector,

Index TermsEvent pricing, machine learning, dynamic pric- ing, regression, event management, vendor services.

1. INTRODUCTION

   Event management pricing is often manual and based on experience. Vendors tend to fix prices based on gut feel, or price cards, without acknowledging that demand does vary by season, type of event, and customer wants. Dynamic pricing used in other industries such as travel and e-commerce can change the event services industry by adjusting prices in real- time or periodically based on multiple influencing factors.

   When it comes to price flexibility, machine learning (ML) has the power to help us build intelligent pricing models; and event services platforms can offer smarter price recom- mendations for vendors, with the help of historical vendor data and predictive algorithms, while also increasing vendor profitability, and making customers happy.

2. LITERATURE REVIEW

   Dynamic pricing has become a key strategy in service- oriented platforms, allowing businesses to respond more effectively to market shifts, customer behavior, and time- sensitive demands. With the emergence of Artificial Intelli- gence (AI) and Machine Learning (ML), pricing strategies have evolved beyond static models, enabling intelligent and real-time decision-making.

   1. Reinforcement Learning for Dynamic Pricing in E- Commerce

      Liu et al. (2019) introduced a deep reinforcement learning (DRL) model tailored for dynamic pricing on Alibabas Tmall platform. The pricing process was modeled as a Markov Decision Process (MDP), where the reward was based on changes in revenue conversion rates. Experimental validation, including both offline data and live trials involving a wide range of toy products, demonstrated an increase of up to 11% in conversion rates compared toional expert-driven pricing methods.

   2. AI-Based Dynamic Pricing and Revenue Optimization

      Thota et al. (2024) outlined a three-tier framework leverag- ing AI for pricing decisions. The model starts by segmenting customers using K-Means clustering, followed by the appli- cation of multiple linear regression to estimate price ranges. Logistic regression is used in the final stage to determine the likelihood of purchase. The proposed approach achieved an impressive 94% accuracy in identifying price sensitivity segments and led to a 15% boost in revenue compared to standard techniques.

   3. AI Tactics for Real-Time Dynamic Pricing

      Building upon existing frameworks, further research has explored real-time pricing adaptations based on customer inter- action patterns such as visit frequency and booking behavior.

      These adaptive systems demonstrated strong performance with 92% accuracy in estimating price ranges and an 18% increase in booking success rates, which proved to be effective in fast- evolving demand environments.

   4. Fairness-Aware Reinforcement Learning in Pricing

   Maestre et al. (2018) proposed a reinforcement learning approach that integrated fairness constraints into pricing mech- anisms. The model focused on fair pricing among different customer groups, effectively lowering instances of biased pricing by over 80%. Importantly, this fairness was achieved with only minimal impact on overall profit margins.

3. METHODOLOGY

   This section presents the end-to-end methodology used to design, develop, and evaluate the dynamic price prediction system for event service vendors. The process includes data collection, description, pre-processing, model selection, train- ing, and performance evaluation.

   1. Problem Definition

      Event service providers often rely on static or intuition- based pricing, which does not account for market fluctuations, seasonal demand, and contextual features such as event type or location. This results in inconsistent pricing, missed revenue opportunities, and reduced competitiveness. The proposed system addresses this by using machine learning models to dynamically predict optimal service prices based on historical and vendor-specific data.

   2. Data Collection

      To build an effective and realistic pricing model, struc- tured data was collected directly from event service providers through a custom-designed Google Form. The form was sent to a diverse group of vendors who offer services such as catering, photography, decoration, and DJ services. The data collected focused on key attributes that significantly influence pricing decisions. These included the type of service (e.g. catering, photography), which determines the nature and complexity of offerings; the base price, representing the vendors starting price point; and the category of events (e.g. wedding, corpo- rate), as different types of events have different budget expec- tations. Additionally, location was captured to account for ge- ographic differences in pricing due to cost of living variations between regions. Seasonality was included to reflect whether the service was being offered during high demand (peak) or low demand (off-peak) periods, significantly affecting price fluctuations. Finally, years of experience were collected to understand how a vendors professional background influences their pricing structure. This data serves as the foundation for training machine learning models capable of generating dynamic and context-aware price recommendations.

   3. Data Description

      The final dataset comprises 312 entries, each representing a unique vendors service profile. The dataset includes both cat- egorical and numerical attributes that influence event service pricing. The features consist of the service type (e.g., catering, photography, decoration), which defines the nature of the offering; the base price, which reflects the vendors initial rate; and the event category (e.g., wedding, corporte, birthday), which helps in understanding demand patterns. Additional categorical variables include location, denoting the region or city where the service is offered, and seasonality, indicating whether the event occurs during peak or off-peak periods. Nu- merical features include years of experience, which signifies the vendors industry exposure; customer rating, an optional metric capturing self-assessed or client-reviewed performance on a 5-point scale; and booking lead time, which represents the number of days in advance the service is typically reserved. The target variable to be predicted is the final service price, influenced by the combination of these input features.

      Fig. 1. vendor dataset used for training and testing the dynamic price prediction model. The dataset includes service type, event category, location, seasonality, vendor experience, and base pricing details..

      Fig. 2. Preview of the event cost dataset used for model training, showing attributes like Month, Function, Location, and Cost.

   4. Data Preprocessing

      There are several steps taken to preprocess the dataset ahead of applying the model.Before feeding the data into the machine learning models, several preprocessing steps were carried out. First, we handled missing valueseither by removing incomplete records or filling them in using statistical methods like the mean or mode, depending on the type of data. Next, we transformed categorical variables such as service type, event category, and location into numerical format using one-hot encoding. This step ensured that each category was represented independently, making it suitable for machine learning algorithms. For numerical features like base price, vendor experience, and booking lead time, we applied Min-Max normalization to scale values between 0 and

      1. This helped prevent any single feature from having too much influence on the model. Finally, the dataset was split into two parts: 80% for training the models and 20% for testing them

      on new, unseen data. This split allowed us to assess how well the models would perform in real-world scenarios.

   5. Model Architecture

      Three regression-based machine learning models were de- veloped and evaluated for their ability to predict dynamic prices. The first model was linear regression which basically provides a baseline model because it models a straight line between the variables and the price output. The second model was Decision Tree Regressor which, like linear regression, will utilize non-linear patterns within the dataset. Decision Tree Regressor is effective on structured datasets that include both categorical and continuous features. The third and most complicated model developed was XGBoost Regressor, which is a gradient boosted ensemble method. XGBoost is considered one of the most robust and accurate ML models available. Each of these models were implemented in Python using the Scikit-learn and XGBoost libraries. The hyperparameter tuning of the Decision Tree and XGBoost models in this chapter was performed using grid search.

   6. Evaluation Metrics

   To measure how well the models predicted prices, we used three commonly accepted regression evaluation metrics. First, the Root Mean Squared Error (RMSE) helped us understand the average size of the prediction errors by calculating the square root of the squared differences between actual and predicted prices. Next, the Mean Absolute Error (MAE) gave a simple average of how far off the predictions were, without considering the direction of the error. Lastly, the R² score, also known as the coefficient of determination, showed how much of the variation in actual prices could be explained by the model. Together, these metrics gave a clear picture of how accurate and reliable each model was when applied to different vendor scenarios.

4. SYSTEM IMPLEMENTATION

   1. Backend and Framework

      Backend and Framework Backend and Framework Python and the Scikit-learn library were utilized to write the machine learning logic necessary for price prediction and model testing. Sophisticated modeling was done through the use of the XGBoost Python API for gradient boosted regressor execu- tion. The entire system is modularly designed with the data preparation pipeline, model inference, and web interface split out to enable scalability. MongoDB was chosen as the backend store because it permits us to store data semi-structured. It was appropriate for semi-structured data (e.g. vendor profiles, prediction history and price histories). MongoDB is schema- less and permits fields and service types to be implemented without the constraints of table organization.

      The web interface was designed to support two potential web frameworks (Flask ) for deployment. Flask is used at prototyping and development phases as it is lightweight and can directly consume Python-based ML models. The architec- ture also supports API status endpoints for submitting forms.

      Fig. 3. System Architecture for Dynamic Price Prediction. The architecture illustrates how vendor input flows through the web interface, backend, and machine learning engine to generate cost predictions using an SQLite database and seasonal logic.

   2. System Architecture

      Our system is designed to predict event service prices using machine learning. Vendors provide input such as event type, location, and season through a simple form. This data is processed by the backend, where its cleaned and converted into a format the model can use.

      We use a Linear Regression model that takes these inputs and predicts an estimated cost. The backend then sends this predicted cost back to the vendor interface. All data is stored in an SQLite database, which helps both in storing records and improving the model over time.

      This setup allows smooth communication between the user interface, backend server, machine learning engine, and database to generate fast and accurate pricing.

   3. Output

      The system geneBased on details like service type, event category, vendor experience, location, and seasonal factors, the system processes the input using trained models to generate a price estimate. Once a prediction is generated, it is provided to the vendor or system admin as a real-time price prediction via the web interface or API response. In addition to numeric prediction, a feature importance visual representation of the prediction is produced to indicate the relative importance of factors impacting the predicted price; this is beneficial for vendors to understand the underlying relative weights attributed to the event type, seasonality, or years of experience in optimal pricing decisions.

      Feature importance is calculated using model-specific tech- niques. In the case of Decision Tree and XGBoost, feature importance is calculated as an information gain when splitting the feature during training. The visual results are provided as a bar chart, and show the most important contributing factors with their respective feature importance scores. Because of the interpretability component of the decision tree, it builds confidence in the vendors trust of the AI system, to understand

      the transparency in how pricing decisions are derived. The prediction results can also be saved to the MongoDB database, for later analysis or review, or as data for automated reporting and pricing dashboards. Future updates will include real-time alerts for predicted increases in prices.

5. RESULTS AND DISCUSSION

   Out of the three models that were run, XGBoost Regres- sor performed the best with lowest RMSE and highest R² value and is therefore the most suitable model to use for dynamic price forecasting. As compared to Linear Regression and Decision Tree Regressor, they had higher margins of error and lower accuracy, particularly on non-linear feature combinations. Feature importance analysis also indicated that even type, seasonality, and vendor experience were the most influential factors on price predictions. Vendors providing ser- vices for wedding events during peak seasons with more years of experience were likely to get higher predicted prices. The model further recognized location as a strong influence factor, particularly for vendors based in urban areas. In comparison to conventional static pricing, the developed system exhibited better accuracy, consistency of prices, and revenue opportunity. Initial testing using test vendor data indicated the ability of the model to enable vendors to better match prices with demand in the market, thus increasing revenue as well as booking opportunities.

   Among models that were experimented with, Linear Regres- sion gave baseline performance for event service price predic- tion. The model was trained on features such as Month and Function type, and tested on a hold-out test set. Performance was evaluated using RMSE and R² metrics.

   As shown in Figure 6, the models predictions show moder- ate correlation with actual prices, with an RMSE of 32,316.99 and an R² score of 0.08.

   Among the models tested, Linear Regression provided baseline performance for predicting event service prices. The model was trained using features like Month and Function type, and evaluated on a hold-out test set. The performance was measured using RMSE and R² metrics.

   As shown in Figure 6, the models predictions show moder- ate correlation with actual prices, with an RMSE of 32,316.99 and an R² score of 0.08.

6. FUTURE WORK

   In an effort to improve the functionality and scalability of the dynamic pricing framework, several enhancements have been identified. First, the framework will be modified to allow for real-time API deployment that enables web and mobile integration for instant price predictions and recommendations. Second, a learning mechanism based on feedback will be inserted into the framework, allowing a model to accumu- late knowledge and learning from vendor feedback, customer reactions, and booking outcomes, and disperse that learning in increasingly accurate predictions. Third, the framework will be enhanced to have regional rates and multiple types of vendors to allow for increased flexibility and situational

   Fig. 4. Actual vs predicted event service prices using Linear Regression.

   Fig. 5. Comparison of actual and predicted event service costs using Linear Regression. The model shows reasonable prediction accuracy across varied inputs.

   awareness in different markets and service categories. Fourth, a vendor focused mobile interface will be designed to easily provide access and price input to the framework for instant predictions on the go, and specifically for small-scale vendors whose primary business is within dynamic environments, who may require the framework to accommodate on-the-go pricing input.

7. CONCLUSION

This paper describes a vendor-centric dynamic pricing sys- tem in the events services using machine learning strategies. By collecting structured data from real vendors, the proposed system uses regression-type models including Linear Regres- sion, Decision Tree, and XGBoost, to produce accurate and dynamic price predictions. The XGBoost model performed best overall in predictions and generalization. The analysis of features indicated that event type, seasonality, and experience were the most important pricing factors. The pricing systems data driven recommendations significantly improved vendor profitability, vendor satisfaction, and inaccuracies associated with traditional static pricing strategies. After planned en- hancements such as no cost integration within an api to collect real-time data trail feedback, learning from user feedback, and the ability to support a mobile interface, the system has

Fig. 6. Google form Screenshot for Vendor Pricing Data Collection

Fig. 7. Average event cost by season showing that winter events tend to be the most expensive, suggesting a seasonal pricing trend.

significant prospects to deploy at scale in the evolving event management ecosystem.

REFERENCES

1. J. Liu, J. Ni, and H. Xu, Dynamic Pricing on E- Commerce Platform with Deep Reinforcement Learning: A Field Experiment, SSRN, 2022. [Online]. Available: https://doi.org/10.2139/ssrn.4051563

2. V. Thota and R. Sharma, AI-Based Dynamic Pricing and Revenue Optimization Strategies in E-Commerce, Migration Letters, vol. 21, no. 1, pp. 101117, 2024.

   Fig. 8. Correlation matrix between month, function type, and event cost. A weak negative correlation is observed between function type and cost.

3. A. Shah and K. Jain, E-Commerce Management and AI-Based Dynamic Pricing Revenue Optimization Strategies, Journal of Retailing and Consumer Services, vol. 59, p. 102422, 2021.

4. R. Maestre and M. Lo´pez-Sa´nchez, Fairness-Aware Reinforcement Learning for Pricing in E-Commerce Systems, in Proceedings of the AAAI Conference on Artificial Intelligence, 2018, pp. 14891495.

5. L. Xie, H. Liu, and Y. Wang, Dynamic Pricing and Energy Consumption Scheduling with Reinforcement Learning, IEEE Transactions on Smart Grid, vol. 11, no. 4, pp. 34253435, Jul. 2020.

6. J. Liu et al., Dynamic Pricing on E-Commerce Plat- forms with Deep Reinforcement Learning: A Field Experiment, Proceedings of the AAAI Conference on Artificial Intelligence, vol. 35, no. 3, pp. 23342342, 2021.