Machine Learning Techniques to Assess Faceted Blue Sapphire Gemstone Prices

The gem industry is a giant pillar that supports the economic growth of Sri Lanka. Gemstone market prices rely on the implicit knowledge of professional gem evaluators. Complexity in gemstone valuation due to their inherent variability leads to subjective decisions. This creates a necessity for establishing a standardized model that mitigates the risk of overpricing or underpricing of gemstones. Developing a standardized model for gemstone valuation benefits stakeholders and upholds the esteemed reputation of the gem industry. This research developed high-tech models to assess faceted blue sapphires, taking the exported gemstone records from the National Gem and Jewellery Authority. The dataset comprehends a total of 881 records with both single (409) and batch (472) assessments of faceted blue sapphire gemstones. This study identified the pivotal factors for gem valuation, such as colour, inclusions, cracks, cut, weight, state of treatment, and calibration. Multiple machine learning regression models, including multiple linear regression (MLR), quantile regression (QR), support vector regression (SVR), feedforward neural network (FFNN), and generalized regression neural network (GRNN), were employed in building pricing models. The MLR and QR models resulted in inadequate pricing models due to the violation of model assumptions. While SVR, FFNN, and GRNN exhibited similar performances, GRNN outperformed in evaluating both single and batch assessments. For pricing single assessments, the SVR model with the Gaussian kernel exhibited a superior performance. The proposed efficient and reliable pricing models to assess faceted blue sapphires will directly benefit appraisers and indirectly gem collectors and investors. This study also recommends exploring future research paths, including the integration of decision support tools like computer-trained pricing models, to assess other varieties of gemstones and enhance their broad applicability.