Recommendation system for horticultural commodities

Abstract

Horticultural commodities commonly have fluctuating prices due to their nature. Seasonality and climate are the main factors that cause their prices to fluctuate. Price instability causes a planning on horticultural cultivation to become difficult. Local farmers would intuitively know the planning by their experience, but this might be too complicated for those farmers who are new or have no experience. Thus, the proposed recommendation system would be able to help the new farmers to set a schedule for horticultural cultivation. The proposed recommendation system consists of three phases: price prediction, commodity recommendation, and cultivation scheduling. A hybrid of Long Short-Term Memory Neural Network (LSTM) with Seasonal and Trend Decomposition based on LOESS (STD-LOESS) are used as price prediction model. The proposed model can provide multistep price prediction with acceptable accuracy. The predicted prices are used with the preferred cultivation period to find the most suitable commodities to be cultivated for that period. Finally, the cultivation schedule with the best starting time and harvesting time for suitable commodities based on seasonality, price, cultivation location and production index is returned as the result, thus farmers would be able to decide when to start the cultivation and when to harvest the commodities.