[ ko ] | [ en ]

Structured Data Analysis – Advanced League

11th Big Contest 2023

Optimizing Classical Concert Pricing at Seoul Arts Center

🎻

Miraclassic de BOAZ
Jaewook Shin	Hyeyeon Kim	Kyuyeon Park	Yujin Choi

 

🏆Grand Prize🏆

(Minister of Science and ICT Award)

 
 

🎯 Project Goal

1. Design a pricing model that accommodates a diverse range of audience needs,

   • Addressing the limitations of the current pricing system, which overlooks varying interests and preferences
   • Proposing new seat grouping standards based on audience profiles and concert characteristics
   • Enhancing accessibility and enjoyment for all concertgoers

2. Ensure transparency in pricing decisions through data-driven justification,

   • Providing logical, evidence-based pricing grounds
   • Making the pricing strategy more open and trustworthy
   • Empowering audiences with clear and reliable pricing information

🧠 Summary

• Data preprocessing & EDA
• Audience Seat Preference Patterns by Genre
• Demand Prediction Modeling
• Price Modeling

1️⃣ Audience Seat Preference Patterns by Genre

We hypothesized that seat preferences would vary by genre, so we split the data by genre and performed K-Means clustering on seat-level data within each group.

To determine the optimal number of clusters, we analyzed the Elbow Curve along with the average and distribution of silhouette coefficients.

• Left (Symphony): Preference for Block D over Block B (left side), possibly due to the orchestra layout where soloists face the conductor’s left
• Middle (Classical): Slight preference for central seating compared to symphony; extreme side seats may distort balance between instruments
• Right (Choral): Strong preference for broad central zones, offering balanced audio from diverse vocal sections

2️⃣ Demand Prediction Modeling

We performed two separate regression tasks:

1. R: General reservation rate
2. M: Membership conversion rate

🎯 Input Features:

• Performance start time
• Pre-sale status
• Genre
• International artist (yes/no)
• Running time
• Day of week
• Month

⚙️ Model Training:

• Data split: 80% train / 20% test
• Cross-validation: 5-fold CV during GridSearch
• Evaluation metrics: MSE, MLSE

📊 Model Performance Comparison:

CatBoost showed the lowest MSE, so it was selected as the final model for predicting general reservation rate (R).

Although MLP achieved the lowest MSE for membership conversion rate (M), CatBoost was chosen for its better interpretability.

Comparison of MSE and MLSE scores across 6 regression models.

SHAP value analysis was conducted after model training to extract feature-level insights.

3️⃣ Price Modeling

1. Cluster performances based on characteristics

K-Means clustering on performance features resulted in 6 clusters, validated through silhouette score analysis.

2. Set base price using the average price within each cluster

The base prices for the highest and lowest seat grades were set using the average of the highest and lowest prices within each assigned cluster.

3. Adjust base price using predicted reservation rate and membership conversion rate

The base prices were adjusted using a weighted function.

4. Distribute final prices across seat grades within the [min–max] range

🏁 Conclusion

✅ Practical Applications

• Automated seat grouping and price estimation for new performances based on input characteristics

• Enhanced seat selection interface that displays additional insights, such as how many customers with certain profiles previously reserved a given seat grade

• Transparent pricing rationale, helping to build audience trust

• Data-driven guidance to support more informed purchase decisions

✅ Expected Benefits by Stakeholder

🎭 Seoul Arts Center (Venue Operator)

• Enables optimal pricing by adjusting prices based on predicted demand, maximizing both attendance and revenue
• Improves customer satisfaction through reasonable pricing and clearer seat categorization, encouraging repeat visits and revitalizing interest in classical concerts

👥 Audience

• Provides a clear guideline for selecting appropriate seats based on individual interest in classical performances
• Helps maximize consumer utility within each individual's willingness to pay

🔧 Tools & Techniques

• Python (pandas, sklearn, matplotlib)
• Linear Regression, KMeans Clustering

📁 Included

• /bigcontest_2023_pitch.pdf – Full presentation slides

---

Copyright 2023. Miraclassic de BOAZ All rights reserved.