Market Demand-Driven Forecasting of Diamond Prices using ARIMA and LSTM Approaches

Accurate prediction of diamond prices is vital for consumers, retailers, and investors. However, price fluctuations are influenced not only by diamond features such as carat, cut, and clarity but also by external market dynamics and seasonal demand patterns. In this study, we forecast diamond prices using two prominent techniques: the AutoRegressive Integrated Moving Average (ARIMA) model and Long Short-Term Memory (LSTM) neural networks. The dataset, originally sourced from Kaggle, lacked a temporal structure. To overcome this, we introduced a synthetic time series by adding demand-driven labels and assigning dates based on seasonal price behavior. To reflect actual market behavior, high-value diamonds were associated with peak demand periods, particularly between November and February. For forecasting purposes, both ARIMA and LSTM models were applied to predict diamond prices for the upcoming January and the subsequent year. Their performance was compared using 2017 data as a baseline. The ARIMA model, configured with parameters (p=9, d=0, q=1), demonstrated superior accuracy over the LSTM model with a time step of 30. Specifically, ARIMA achieved lower error metrics, with a Mean Absolute Error (MAE) of 4245.53 and a Root Mean Squared Error (RMSE) of 4985.35, compared to the LSTM’s MAE of 4907.62 and RMSE of 5561.07. These findings suggest that ARIMA is more effective for short-term diamond price forecasting in this context.