An Interpretable Produce Price Forecasting System for Small and Marginal Farmers in India using Collaborative Filtering and Adaptive Nearest Neighbors

Abstract: Small and marginal farmers, who account for over 80% of India's agricultural population, often sell their harvest at low, unfavorable prices before spoilage. These farmers often lack access to either cold storage or market forecasts. In particular, by having access to cold storage, farmers can store their produce for longer and thus have more flexibility as to when they should sell their harvest by. Meanwhile, by having access to market forecasts, farmers can more easily identify which markets to sell at and when. While affordable cold storage solutions have become more widely available, there has been less work on produce price forecasting. A key challenge is that in many regions of India, predominantly in rural and remote areas, we have either very limited or no produce pricing data available from public online sources. In this paper, we present a produce price forecasting system that pulls data from the Indian Ministry of Agriculture and Farmers Welfare's website Agmarknet, trains a model of prices using over a thousand markets, and displays interpretable price forecasts in a web application viewable from a mobile phone. Due to the pricing data being extremely sparse, our method first imputes missing entries using collaborative filtering to obtain a dense dataset. Using this imputed dense dataset, we then train a decision-tree-based classifier to predict produce prices at different markets. In terms of interpretability, we display the most relevant historical pricing data that drive each forecasted price, where we take advantage of the fact that a wide family of decision-tree-based ensemble learning methods are adaptive nearest neighbor methods. We show how to construct heuristic price uncertainty intervals based on nearest neighbors. We validate forecast accuracy on data from Agmarknet and a small field survey of a few markets in Odisha.