Forecasting Prices of Black Pepper in Kerala and Karnataka using Univariate and Multivariate Recurrent Neural Networks Kiranjith K A 1 and Pramila R M 2 1-2 Christ (Deemed to be University), Pune Lavasa Campus, Maharashtra, India kiranjith.a@science.christuniversity.in, pramila.rm@christuniversity.in Abstract—Our country has a high level of agricultural employment. Price swings harm the economy of our country. To combat this impact, forecasting the selling price of agricultural products has become a need. Forecasts of agricultural prices assist farmers, government officials, businesses, central banks, policymakers, and consumers. Price prediction can then assist in making better selections in this area. Black pepper, sometimes known as the "King of Spices," is a popular spice farmed and exported in India. The largest producers of black pepper are Karnataka and Kerala. For black pepper in Kerala and Karnataka, this study provides a univariate and multivariate price prediction model using Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU). The data is denoised using Singular Spectral Analysis (SSA). The most accurate method is the multivariate variate LSTM technique, which uses macroeconomic variables. It has a Mean Absolute Percentage Error (MAPE) of 0.012 and 0.040 for Kerala and Karnataka, respectively. Index Terms— Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Singular Spectral Analysis (SSA), Black Pepper, Macroeconomic, Time series. I. INTRODUCTION The price of black pepper, a trade-dependent commodity, fluctuates a lot. The prices of black pepper exhibit nonlinear patterns. Because of the greater connection with the foreign markets, price volatility might be directly conveyed to domestic pricing. Agriculture analytics is projected to expand in the modern era. Big data and artificial intelligence are two instances of extensively used technologies. These instruments can be used to diagnose crop diseases, manage crops, control pests, monitor disease transmission, monitor irrigation, control soil nutrients, monitor production, and ensure the preservation of agricultural products. Several statistical methods are used in traditional time series prediction methodologies, including auto-regressive integrated moving averages (ARIMA) and seasonal auto-regressive integrated moving averages (SARIMA). It has been proven that these tactics work. These methods can be used only when the data has been transformed to stationary. RNNs are used to reveal and forecast volatile nonlinear data. For the Price Prediction of Black Pepper in Karnataka, univariate techniques with high accuracy have already been developed. In this paper, univariate models are implemented using RNN, and macroeconomic variables are included as part of the multivariate model for price prediction in prominent black pepper-producing states; Kerala and Karnataka. SSA has also been used to denoise the data. Grenze ID: 01.GIJET.8.2.522 © Grenze Scientific Society, 2022 Grenze International Journal of Engineering and Technology, June Issue