Intelligent Price Forecasting System for Spice Traders with Machine Learning

Abstract: Spices are essential agricultural products that hold considerable economic importance and fulfill various roles in culinary, medicinal, and industrial fields. Black pepper, a spice traded worldwide, experiences frequent price changes due to seasonal variations, inconsistent quality, and disruptions in the supply chain. This research introduces a forecasting model aimed at predicting black pepper prices in local markets of Karnataka. Conventional techniques such as Simple Moving Average (SMA), Auto-Regressive Integrated Moving Average (ARIMA), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU) often yield subpar results when faced with irregular data conditions. To overcome this challenge, a Multiple Linear Regression (MLR) model with lag features was developed, utilizing domain-specific feature engineering. The data processing pipeline included steps for managing missing values, outliers, normalization, and identifying temporal patterns. A knowledge-driven nearest neighbor analysis was employed to improve forecasting accuracy. Among all the models assessed, the MLR model recorded the lowest Mean Absolute Percentage Error (MAPE) of 0.22%. The proposed work also features a user interface designed to aid traders in making informed decisions and allows for a more in-depth analysis of black pepper trading trends.