Exportar Publicação

Cardoso, M. G. M. S. & Chambel, L. (2023). Rules for predicting lab-grown diamonds prices. XXVI Conference of the Portuguese Statistical Society.

M. M. Cardoso and L. M. Cardoso,  "Rules for predicting lab-grown diamonds prices", in XXVI Conf. of the Portuguese Statistical Society, Guimarães, 2023

@misc{cardoso2023_1775715623668,
	author = "Cardoso, M. G. M. S. and Chambel, L.",
	title = "Rules for predicting lab-grown diamonds prices",
	year = "2023",
	url = "https://w3.math.uminho.pt/~web/SPE2023/index_en.html#"
}

TY  - CPAPER
TI  - Rules for predicting lab-grown diamonds prices
T2  - XXVI Conference of the Portuguese Statistical Society
AU  - Cardoso, M. G. M. S.
AU  - Chambel, L.
PY  - 2023
CY  - Guimarães
UR  - https://w3.math.uminho.pt/~web/SPE2023/index_en.html#
AB  - Lab-grown gem-quality diamonds have shown fast market-share growth
since the mid-2010s. Several attempts have been made to predict natural diamond
prices based on their characteristics, namely, using Machine Learning techniques.
There are no similar studies referring to lab-grown diamonds. In the present study,
we aim at predicting lab-grown diamond unit prices. Data used were collected
from https://www.1215diamonds.com and https://www.miadonna.com , on 2022,
including 44 443 and 18 283 observations (lab-grown diamonds), respectively. These
data include the diamond prices and also attributes to be used as predictors: Carat
(measured in ct, 0.2 g), Color (7 levels, where the first is the colorless diamond);
Clarity (8 levels where the first has no visible inclusions or internal flaws); Cut,
including 10 different shapes, the predominant being the “Round”. For the task at
hand – predicting lab-grown diamonds prices based on physical characteristics- we
propose using Propositional Rules (RULES). K-Nearest Neighbors (KNN) will be
used as a baseline. We resort to the R packages “FNN” (for KNN) and “Cubist” (for
RULES). Metrics estimated on the test set (30% of the data) including R-Squared,
MAE-Mean Absolute Error, and MAPE- Mean Absolute Percentage Error, enable
evaluation of the predictive capacity of the proposed approach. We conclude that
RULES generally produce better predictions than KNN, and also provide easy-tointerpret
outputs and useful insights regarding specific observations. Future analysis
may include additional predictors such as Cut quality and Certificate
ER  -