![IconGems-Print](images/IconGems-Print.png)
353 – Data Science
Collaborative Cognition for Commodity Price Prediction
Ritwik Chaudhuri
IBM Research India
Manish Kataria
IBM Research India
Ramasuri Narayanam
IBM Research India
Gyana Parija
IBM India Pvt. ltd.
There has been significant interest to predict prices for commodities (raw materials) in spot markets that are volatile as well formulate significant portion of manufacturing costs. This is an important research problem as industries spend several billion dollars in a year to procure such commodities for their business. In this paper, we present a novel approach for price prediction problem by formulating this as collaborative decision-making among autonomous agents or human experts. Following this approach, initially different autonomous agents produce predictions for the commodity leveraging their own knowledge and expertise. Then, these agents collaborate among themselves to share knowledge (full or partial) with each other towards collectively generating prediction of the commodity prices. An agents predictive performance may vary due to changes in the commodity ecosystem as well as various exogenous factors. Since predictions are generated at both an individual agent level as well as at the group level, the groups collective performance is comparatively robust and much better than initial predictions done at an individual agent level. For commodity price prediction problem, where the price is an outcome of multiple competing stakeholders in the market, our proposed collaborative decision-making framework has been observed to be a very powerful methodology in comparison to other state-of-the-art counterparts.