Dynamic Pricing in Electronic Commerce using Neural Network

There exist intelligent agents to aid online sellers to dynamically calculate a competitive price for their products in online markets. However, these intelligent agents usually make a number of assumptions for dynamic pricing. Some intelligent agents assume that sellers consist of prior knowledge about the online market parameters. In other words, the agents assume that the sellers are well aware of other competitors' pricing strategies, consumers purchase preferences, consumers' reservation price, profit made by other competing sellers etc. In addition, other agents assume that price is the only attribute that determines consumers' purchase decision. On the contrary, in real life sellers have limited or no prior knowledge about the market parameters. In addition, nowadays along with price other attributes such as after sale service, product quality etc. contribute in determining consumers' purchase decision. In this thesis, we propose an approach where sellers have limited knowledge on market parameters. We also assume that buyers' purchase decisions are based on multiple attributes. We are using a feed-forward neural network approach for calculating a competitive price dynamically to increase the sellers' revenue. Product price, product quality, delivery time, after sales service and seller's reputation are taken into consideration while determining the competitive price of the product by our model. In our experimental evaluation we showed that once the sellers, by considering the five attributes, set an initial price of the product, our model adjusts the price of the product automatically with the help of neural network in order to raise the revenue. In setting the initial price of a product, we assume that sellers use their prior knowledge about the prices of the product offered by other competing sellers. Any other prior knowledge like buyer demand or competitor's price setting behaviors is not used in our evaluation. The experimental results portray the effect of considering the five attributes in earning revenue by the sellers. Before concluding with directions for future works, we discuss the value of our approach in contrast with related work.