Enhancing Scent Selection with Machine Learning

• Synesthesia: a neurological phenomenon where one sensory stimulus involuntarily triggers another perception.
• Color Psychology: the study of how colors influence human emotions, perceptions, and behavior.
• Fragrance Expertise: combining these principles to design a system that connects visual, textual, and olfactory signals into meaningful recommendations.

Finding the right fragrance is highly personal — it reflects identity, emotions, and moments. This system acts as an expert advisor, helping users discover scents that match both mood and style.

1. Data Analysis: Explore historical fragrance data to understand key industry trends and most used ingredients.
2. Recommendation Engine: Use AI to provide personalized fragrance recommendations.
3. Fashion Integration: Link fragrances with fashion and lifestyle to enhance self-identity.

• Source: Fragrantica.com (via Kaggle)
• Size: 24,063 rows × 18 columns
• Features: brand, country, year, gender, ratings, olfactive notes (top, middle, base), perfumer, accords, family.
• +1600 fragrance notes across 20 olfactive families.

1. Data Cleaning & Exploration
   • Standardization of fragrance dataset.
2. Textual Processing
   • Descriptions vectorized with TF-IDF.
   • Similarity computed using KNN.
   • Query expansion with synonyms (WordNet).
3. Image Processing
   • Background removal (rembg).
   • Face removal (OpenCV Haar Cascade).
   • Dominant color extraction (KMeans).
   • Color-to-tag mapping (e.g., blue → aquatic).
4. Recommendation Engine
   • Combines text-based and image-based signals.
   • Provides personalized fragrance suggestions.

• Image-Based Recommender: Maps uploaded images and palettes to fragrance families.
• Text-Based Recommender: Suggests similar perfumes based on olfactive descriptions.
• Hybrid Model: Integrates image + text to improve precision and personalization.

• Language: Python
• Data Processing: Pandas, Matplotlib
• Image Processing: OpenCV, PIL, rembg, scikit-learn (KMeans)
• Text Processing: NLTK, TF-IDF, Nearest Neighbors
• Deployment (optional): Streamlit

.
├── data/
│   ├── fragrance_ML_model.csv
│   ├── fragrance_database.csv
│   └── colors.csv
├── fragrance_code/
│   ├── data_loader.py
│   ├── image_processing.py
│   ├── processing_text.py
│   ├── model_text.py
│   ├── recommender_image_based.py
│   └── recommender_text_based.py
├── notebooks/
│   └── fragrance_EDA.ipynb
├── models/
│   ├── tfidf_knn_model.pkl
│   ├── vectorizer.pkl
│   └── ...
└── README.md