**2. Related work**

264 Bio-Inspired Computational Algorithms and Their Applications

in the services for procuring goods and items either for commercialize purposed or for personal used. Online auctions have been reported as one of the most popular and effective ways of trading goods over the Internet (Bapna *et al.* 2001). Electronic devices, books, computer software, and hardware are among the thousands items sold in the online auctions every day. To date, there are 2557 auction houses that conduct online auctions as listed on the Internet (Internet Auction List, 2011). These auction houses conduct different types of auctions according to a variety of rules and protocols. eBay, as one of the largest auction house alone has more than 94 million registered users and had transacted more than USD 92 billion worth of goods during 2010 (eBay, 2010). These figures clearly show the importance of online auctions as an essential method for procuring goods in today's e-

The auction environment is highly dynamic in nature. Since there are a large number of online auction sites that can be readily accessed, bidders are not constrained to participate in only one auction; they can bid across several alternative auctions for the same good simultaneously. As the number of auction increases, difficulties such as monitoring the process of auction, tracking bid and bidding in multiple auctions arise when the number of auctions increases. The user needs to monitor many auctions sites, pick the right auction to participate, and make the right bid in order to have the desired item. All of these tasks are somewhat complex and time consuming. The task gets even more complicated when there are different start and end times and when the auctions employ different protocols. For this reasons, a variety of software support tools are provided either by the online auction hosts or by third parties that can be used to assist consumers

The software tools include automated bidding software, bid sniping software, and auction search engines. Automated bidding software or proxy bidders act on the bidder's behalf and place bids according to a strict set of rules and predefined parameters. Bid sniping software, on the other hand, is a practice of placing of bid a few minutes or seconds before an auction closes. These kinds of software, however, have some shortcomings. Firstly, they are only available for an auction with a particular protocol. Secondly, they can only remain in the same auction site and will not move to other auction sites. Lastly, they still need the intervention of the user, that is, the user still needs to make decision on the starting bid

commerce market.

Begin

End Fig. 1. The structure of a Genetic Algorithm

when bidding in online auctions.

(initially) and the bid increments.

Generation = 0

Randomly Initialize Population While termination criteria are not met Evaluate Population Fitness

 Crossover Process Mutation Process Select new population Generation = Generation + 1

> Genetic algorithm has shown to perform well in the complex system by which the old search algorithm has been solved. This is due to the nature of the algorithms that is able to discover optimal areas in a large search space with little priori information. Many researches in auctions have used genetic algorithm to design or enhance the auction's bidding strategies. The following section discusses works related to evolving bidding strategies.

> An evolutionary approach was proposed by Babanov (2003) to study the interaction of strategic agents with the electronic marketplace. This work describes the agents' strategies based on different methodologies that employ incompatible rules in collecting information and reproduction. This work used the information collected from the evolutionary framework for economic studies as many researches have attempted to use evolutionary frameworks for economics studies (Nelson, 1995; Epstein & Axtell, 1996; Roth, 2002; Tesfatsion, 2002). This evolutionary approach allows the strategies to be heterogeneous rather than homogenous since only a particular evolutionary approach is applied. This work has shown that the heterogeneous strategies evolved from this framework can be used as a useful research data.

> ZIP, introduced by Cliff, is an artificial trading agent that uses simple machine learning to adapt and operate as buyers or sellers in online open-outcry auction market environments (Cliff, 1997). The market environments are similar to those used in Smith's (Smith, 1962)

Performance of Varying Genetic Algorithm Techniques in Online Auction 267

The bidding algorithm for this framework is shown in Fig. 3. Let *Item\_NA* be a boolean flag to indicate whether the target item has already been purchased by the agent. Assume that the value of *pr* is based on the current reliable market prices observed from past auctions and that the marketplace is offering the item which the agent is interested in. While the bidder agent has not obtained the desired item, the bidder agent needs to build an active auctions list in order to keep track of the current active auction. Active auction is defined as

auction that is ongoing or just started but has not reach the ending time yet.

 *Select target auction as one that maximizes agent's expected utility;* 

 *if* ((t ≥ *σi* ) *and* (t ≤ *ηi* ) *or* (*Si* (*t*) = *ongoing*)

 *Bid in the target auction using current maximum bid as reservation price at this time;* 

In order to build the active auction list, the bidder agent follows the algorithm as shown in Fig. 4. *Si* (*t*) is a boolean flag representing the status of auction *i* at time *t*, such that *i є A* and *Si* (*t*) *є (ongoing; completed)*. Each auction *i є A*, has a starting time *σi*, and its own ending time *ηi*. The active auction list is built by taking all the auctions that are currently running at time *t*. In English and Vickrey auctions, any auction that has started but has not reached its ending time is considered as active. *Si* (*t*) is used in Dutch auctions since the ending time of

 *Calculate current maximum bid using the agent's strategy; Select potential auctions to bid in, from active auction list;* 

Fig. 2. The Marketplace Simulator

*while (t ≤ tmax ) and (Item\_NA = true)* 

 *Build active auction list;* 

Fig. 3. The bidding agent's algorithm

*for all i є A* 

 *endif endfor*

this type of auction is not fixed.

 *add i to L*(*t*)

Fig. 4. Building active auction list algorithms

*Endwhile*

**3.2 Bidding strategy** 

experimental economics studies of the CDA and other auction mechanisms. The aim of each zip agent is to maximize the profit generated by trading in the market. A standard genetic algorithm is then applied to optimize the values of the eight parameters governing the behavior of the ZIP traders which previously must be set manually. The result showed that GA-optimized traders performed better than those populated by ZIP traders with manually set parameter values (Cliff, 1998a; Cliff, 1998b). This work is then extended to 60 parameters to be set correctly. The experiment showed promising result when compared to the ZIP traders with eight parameters (Cliff, 2006). Genetic algorithm is also used to optimize the auction market parameters setting. Many tests have been conducted on ZIP to improve the agent traders and the auction market mechanism using genetic algorithm (Cliff, 2002a; Cliff, 2002b). Thus, ZIP was able to demonstrate that genetic algorithm can perform well in evolving the parameters of bidding agents and the strategies.

In another investigation, a UDA (utility-based double auction) mechanism is presented (Choi et, al. 2008). In UDA, a flexible synchronous double auction is implemented where the auctioneer maximizes all traders' diverse and complex utility functions through optimization modeling based on genetic algorithm. It is a double auction mechanism based on dynamic utility function integrating the notion of utility function and genetic algorithm. The GA-optimizer is used to maximize total utility function, composed of all participants' dynamic utility functions, and matches the buyers and sellers. Based on the experimental result, it performance is better than a conventional double auction.
