This project explores the relationship between a perfume's fragrance notes and the subjective experience of its emergent accords using a variety of statistical methods including mutual information, principal component analysis, hierarchical agglomerative clustering, and multiclass/multilabel classification. Given a perfume’s list of notes and concentrations, can the resulting accords be predicted? Can the target consumer (male, female, or both) be derived from the notes and accords? How does a perfume’s makeup influence its popularity? For example, if Channel were to create a new perfume targeted to men, which notes/accords should be selected to maximize sales?
The final write-up is included in:
- report.pdf
The Python code is broken out into three Jupyter notebooks focusing on data cleaning, exploratory analysis, and model fitting:
- code/data_cleaning.ipynb
- code/exploratory_analysis.ipynb
- code/model_fitting.ipynb
The data set was sourced from https://www.kaggle.com/sagikeren88/fragrances-and-perfumes, though it appears to no longer be available.