FlexiTicket

Neural Network Dynamic Pricing System for Cinema Attendance Optimization

A Multi-Layer Perceptron (MLP) that uses cross-entropy loss for intelligent price calculations in cinema dynamic pricing systems, designed to maximize attendance while optimizing revenue through strategic price reductions.

Project Overview

FlexiTicket addresses the critical challenge of cinema attendance optimization by implementing a neural network that predicts optimal price reduction strategies. The system analyzes multiple factors including historical attendance, public expectations, and time-based patterns to recommend price adjustments that increase accessibility while maintaining profitability.

Key Objectives

• Increase Cinema Accessibility: Make movies more affordable through intelligent pricing
• Maximize Attendance: Fill more seats through strategic price reductions
• Revenue Optimization: Balance reduced prices with increased volume
• Data-Driven Decisions: Replace guesswork with ML-powered insights

Model Architecture

Neural Network Structure

Input Layer (5 neurons)
    ↓
Hidden Layer 1 (16 neurons) + ReLU
    ↓
Hidden Layer 2 (16 neurons) + ReLU
    ↓
Output Layer (12 neurons) + Softmax

Total Parameters: ~500 trainable parameters Loss Function: Cross-Entropy Loss Activation: ReLU (hidden layers), Softmax (output)

Input Features

The model takes 5 normalized input features:

Feature	Type	Range	Description
Price Reduction Class	Integer	0-11	Historical price reduction category applied
Actual Attendance	Float	0-1	Normalized actual attendance rate
Expected Attendance	Float	0-1	Normalized predicted attendance without intervention
Public Expectations	Float	0-1	Sentiment/expectation score for movie block
Time Block	Integer	0-9	Time slot identifier (e.g., morning, afternoon, evening)

Feature Engineering Notes

• Attendance values are normalized by theater capacity
• Public expectations derived from reviews, ratings, and social media sentiment
• Time blocks represent standardized time periods (e.g., 0=early morning, 9=late night)

Output Classifications

The model outputs 12 price reduction categories:

Class	Reduction %	Strategy
0	No reduction (0%)	Premium pricing
1	5% reduction	Minimal discount
2	10% reduction	Light discount
3	15% reduction	Standard discount
4	20% reduction	Moderate discount
5	25% reduction	Significant discount
6	30% reduction	High discount
7	35% reduction	Major discount
8	40% reduction	Deep discount
9	45% reduction	Maximum discount
10	50% reduction	Half-price special
11	55%+ reduction	Emergency fill strategy

Performance Metrics

• Primary: Cross-Entropy Loss (minimization)
• Secondary: Classification Accuracy
• Business: Attendance Lift % and Revenue Impact

Expected Outcomes

Business Impact

• Attendance Increase: 15-30% average increase in occupied seats
• Revenue Optimization: Maintain 85-95% of full-price revenue through volume
• Customer Satisfaction: Increased accessibility and perceived value
• Market Share: Competitive advantage through dynamic pricing

Model Performance Targets

• Training Accuracy: >85%
• Validation Accuracy: >80%
• Cross-Entropy Loss: <0.5
• Convergence: Within 50-100 epochs

Business Logic Validation

• Higher public expectations → Lower price reductions needed
• Large attendance gaps → Higher price reductions recommended
• Peak time blocks → More conservative pricing
• Off-peak periods → More aggressive discounting