This project focuses on analyzing perfume composition and popularity using the Fragrantica Fragrance Dataset as the primary dataset. It contains around 20,000 perfumes and includes detailed information such as fragrance notes (top, middle, base), accords, gender category, and user ratings. With 18 structured columns, this dataset provides a rich representation of olfactory composition, making it particularly suitable for analyzing relationships between scent structure and perceived attractiveness.
To incorporate economic and market-related aspects, we use the Perfume E-commerce Dataset 2024, which contains around 2,000 perfumes listed on eBay. This dataset includes pricing and product listing information, allowing us to extend the analysis to market positioning. However, since the data originates from eBay, it primarily reflects markets where eBay is widely used, particularly the United States. Therefore, price-related insights should be interpreted within this geographic context.
The overall data quality is good, as both datasets are structured and publicly available. However, preprocessing is required. Fragrance notes are stored as text and must be parsed into structured formats. Similar notes need to be grouped into broader categories (e.g., citrus, woody, floral), inconsistencies in naming must be resolved, and missing values (especially in ratings or prices) will be handled. Overall, preprocessing remains moderate and consistent with the course requirements.
This project aims to answer the following question: what makes a perfume attractive, and how does its composition influence its popularity and market value? More specifically, we investigate whether certain fragrance notes or combinations of notes are systematically associated with higher user ratings and higher prices.
The main objective is to identify patterns in olfactory composition that explain why some perfumes are more appreciated than others. In particular, we compare perfumes marketed toward men and women to determine whether preferences in scent composition differ. In addition, we analyze how perfume composition evolves over time using release year data, allowing us to identify trends in olfactory preferences.
The available data does not allow for direct seasonal analysis (e.g., summer vs winter). Instead, temporal analysis at the level of release year provides a more reliable proxy for studying trends.
The goal is to produce clear and intuitive visualization on perfume data including preferences depending gender, year and tone.
The Fragrantica dataset contains 24,063 perfumes across 18 columns. The gender split is 47.3% women, 32.0% unisex, and 20.7% men. Missing values are limited to a few columns: release year is absent for 8.5% of entries, and 94.4% of perfumes list only one perfumer (the second perfumer field is intentionally sparse).
Fragrance notes in the Top, Middle, and Base columns are present for the vast majority of entries. User ratings, stored as a numeric score between 3.0 and 4.0, are available for a small subset of perfumes (mean = 3.99, std = 0.10), suggesting that only popular or well-reviewed fragrances accumulate enough votes to appear.
Across all three note layers (top, middle, base), the most frequent ingredients are musk (11,022 occurrences), bergamot (8,617), sandalwood (8,059), jasmine (7,745), and amber (7,730). These five notes dominate the dataset and reflect the broad aromatic vocabulary shared across genders.
Gender-specific preferences are nonetheless visible. For women's perfumes, the most common notes are musk, jasmine, sandalwood, rose, and bergamot — a predominantly floral and soft-woody profile. Men's perfumes favor bergamot, musk, patchouli, amber, and cedar, indicating a shift toward earthier, resinous, and woody compositions. This difference is consistent with established market conventions in perfumery.
The heatmap below expresses each note's share of total occurrences within each gender category. It confirms that jasmine and rose are disproportionately present in women's perfumes, while patchouli and cedar are concentrated in men's. Musk and bergamot stand out as genuinely cross-gender staples.
Regarding ratings, notes with the highest average scores (among those in at least 50 rated perfumes) include grapefruit, musk, amber, pink pepper, lavender, ylang-ylang, and tonka bean, all averaging near 4.0. This suggests that warm bases and fresh citrus top notes correlate with higher user appreciation. The eBay dataset adds a market dimension: men's listings range from $3 to $259 (mean $46) and women's from $2 to $300 (mean $40).
Platforms such as Parfumo Discover provide interactive tools to explore perfumes based on notes, popularity, and categories, illustrating existing approaches to fragrance data visualization.
Some projects have applied machine learning techniques, such as the Perfume Recommender System, which focuses on suggesting fragrances based on similarity rather than interpreting composition.
Other works, such as Scento’s global fragrance map (https://www.scento.com/blog/global-fragrance-map-regional-styles), highlight geographic differences in scent preferences, showing that cultural factors influence fragrance composition. //add scentMap.png
Our approach differs by focusing on interpretable relationships between composition, user ratings, and price. By combining compositional and market data, we aim to uncover patterns in perfume attractiveness, including differences across gender and temporal trends.