We predict lab-grown diamonds’ unit prices based on the traditional 4 Cs—Carat (weight), Colour, Clarity and Cut (shape). For comparative purposes, natural diamond prices are also analysed. The data used originated from online diamond retailers. Supervised learning techniques were primarily selected for their interpretability; however, Random Forests were also included due to their strong performance potential, as highlighted in the literature. The Cubist rule-based algorithm achieved the highest predictive performance on lab-grown diamonds data, while on natural diamonds, it ranked second, following Random Forests. Additional insights were provided by the alternative methods used, including Linear Regression, K-Nearest Neighbours, Regression Trees, and Bayesian Networks. In general, unit prices of lab-grown diamonds have proven much more difficult to predict than those of natural diamonds, where the relationship between the key physical attributes (4 Cs) and prices is more evident. Local interpretability was also explored through two queries, one referring to a good-quality diamond with a standard unit weight (1 carat) and another to a larger (5 carat) high-quality diamond. The Expert’s understanding of these queries provided meaningful contributions regarding the formation of diamonds’ prices. The findings offer valuable insights into diamonds’ price formation and can enlighten consumers and sellers about this constantly evolving market of lab-grown diamonds.

--

Supervised learning,Lab-grown diamonds’ prices,Online diamond market,Interpretable models,Local interpretability

• Matemáticas - Ciências Naturais
• Ciências da Computação e da Informação - Ciências Naturais