Fashion Trend Analysis (DS4420 Final Proj)

The application provides insights into upcoming fashion trends across different categories including styles, colors, patterns, and clothing items.

Features

• 📊 Trend Visualization: Interactive visualization of predicted trends
• 🎨 Color Analysis: Visual grid of trending colors with prediction confidence
• 📈 LSTM Performance: Comparison of predicted vs actual trends
• 🔍 Multi-category Analysis: Insights across styles, patterns, and clothing categories

ML Methodology

Data Collection and Processing

Runway Image Dataset

• Source: Global fashion weeks (last three seasons)
• Collection Method: Web scraping of runway shows
• Data Structure:
  • Designer categorization
  • Seasonal organization
  • Show-specific grouping
• Curation Process:
  • Automated image quality filtering
  • Duplicate removal
  • Metadata standardization

Consumer Trend Data

• Source: Google Trends API
• Time Range: Historical search data spanning multiple seasons
• Features:
  • Weekly search interest metrics
  • Normalized trend values
  • Geographic segmentation

Machine Learning Models

1. CLIP-Based Visual Analysis

• Architecture: Zero-shot classification using CLIP
• Implementation:
  • Multi-label classification for simultaneous attribute detection
  • Confidence threshold: 0.7 for attribute assignment
  • Text prompt engineering for fashion-specific classification
• Categories:
  • Garment types (tops, outerwear, etc.)
  • Patterns (floral, striped, etc.)
  • Colors (with hex codes)
  • Styles (casual, formal, etc.)

2. LSTM Time Series Forecasting

• Architecture:
  • Multiple LSTM layers with dropout
  • Sequence length: Weekly data points
  • Dense output layer for trend prediction
• Training Process:
  • Data normalization
  • Sequence windowing
  • Validation split: 20%
  • Early stopping implementation

ML Results

Visual Analysis Performance

• Overall Accuracy: 87% across all categories
• Category-wise Performance:
  • Garment Classification: 89% accuracy
  • Color Detection: 92% accuracy
  • Pattern Recognition: 85% accuracy
  • Style Attribution: 83% accuracy

Project Structure

.
├── .streamlit/
│   └── config.toml          # Streamlit configuration
├── data/
│   ├── predictions/         # Trend prediction results
│   │   ├── color_predictions.csv
│   │   ├── style_predictions.csv
│   │   ├── category_predictions.csv
│   │   ├── pattern_predictions.csv
│   │   └── super_category_predictions.csv
│   └── lstm_tests/         # Model evaluation results
│       └── *_forecasting_test_results.csv
├── pages/
│   ├── 1_Whats_Trending_Now.py  # Trend visualization page
│   └── 2_LSTM_Test.py           # Model performance page
├── Home.py                 # Dashboard home page
├── requirements.txt        # Python dependencies
└── packages.txt           # System dependencies

Data Files

Prediction Files (data/predictions/)

Required CSV format:

category,predicted,confidence
item1,75.5,High
item2,45.2,Medium
...

LSTM Test Files (data/lstm_tests/)

Required CSV format:

category,actual_fall25,predicted_fall25,error,confidence
item1,70.2,75.5,5.3,High
item2,42.1,45.2,3.1,Medium
...

Setup & Deployment

Local Development

1. Create and activate a virtual environment:

python -m venv venv
source venv/bin/activate  # Unix
venv\Scripts\activate     # Windows

2. Install dependencies:

pip install -r requirements.txt

3. Run the application:

streamlit run Home.py

Cloud Deployment (Streamlit Cloud)

1. Fork/clone this repository
2. Ensure all required data files are present in their respective directories
3. Connect your repository to Streamlit Cloud
4. Deploy using the following settings:
   • Main file path: Home.py
   • Python version: 3.9+
   • Requirements: requirements.txt
   • Additional packages: packages.txt

Dependencies

Python Packages

streamlit==1.31.0
pandas==2.0.3
plotly==5.18.0
numpy==1.21.0
Pillow==9.0.0

System Requirements

python3-dev

Pages

1. What's Trending Now

• Displays top 10 predictions for each category
• Interactive tabs for different trend categories
• Color visualization grid for color trends
• Confidence-based color coding

2. LSTM Test Results

• Actual vs Predicted comparison charts
• Performance metrics (MAE, MSE, Accuracy)
• Detailed comparison tables
• Category-wise analysis

Notes

• All data files must be present in their respective directories for deployment
• CSV files must follow the specified format
• The application is configured for light mode display
• Ensure all paths in .gitignore are properly configured to include data files