Product configuration based on basket analysis of personal computer retail selling.

Fabac, Robert ; Klicek, Bozidar ; Pihir, Igor 等

1. INTRODUCTION

The complexity of modern products, the ever growing offer and diversity of products makes it difficult for the customer to choose. As for ICT products, specifications are rather abstract and it is doubtful whether the customer knows what he has actually chosen when he buys a specific processor and memory, for example. That is why vendors match configurations that are fixed. However, personal preferences of individual buyers are sometimes strong, in a way that, for example, somebody will choose a bigger LCD if this feature is particularly important in his own system of values Our research was conducted in a randomly selected PC shop in a middle-sized Croatian town. We analyzed a sample of approximately 2000 receipt bills issued in the first six months of year 2007, and singled out those that contain Intel and AMD processors.

Earlier research efforts in the domain of customer preferences are numerous. In the domain of financial sector services, segmentation of buyers and the choice criteria used by consumers to choose a home loan (Mylonakis, 2007) was examined by using the method of correlation analysis. There is a number of valuable research examples in the area of conceptual approaches, such as the Customer Relationship Ontology that helps managers formally define which customer segments they want to offer value and through which channels they want to do this (Osterwalder & Pigneur, 2003).

The area of customers' preferences and their decisions in terms of conceptual approach is an issue of interest for many authors (Bachann et al., 2001). The researches and analyses of processors such as those conducted by Raby (2009) represents useful sources of reference for us.

To establish the preferences of those who buy Intel and AMD processors, we chose the methodology of Data Mining which is closely connected with knowledge discovery (Loshin, 2003). Methods of basket analysis and association rule learning are used for the purpose of research. Software tool Statistica in option Association Rules enables usage of three categories for description of the relations: support value, confidence value, correlation value. In practise threshold should be prescribed for three values from the reason of detection of regularities. Under given circumstances, requirements of correlation as one common standard in the research, are substituted with more restrictive option.

2. BUSINESS INTELLIGENCE, DATA MINING AND ASSOCIATION RULE LEARNING

Business intelligence is architecture and collection of integrated operative applications and decision-making support applications which together with data basis enables business groups access toward the relevant data and information (Moss & Atre, 2003). One of the most popular method is data mining, the process of extracting patterns (knowledge) from data..

Patterns of shopping answered the question" What and how often customers buy?" With application of association rule learning method the analysts can discover the rules that prescribe which pairs of articles are purchase together and what are the belonging probabilities (Maimon & Rokach, 2005). This method provides generating of association rules, such as: IF {beer, no bar meal} THEN {crisps}.

3. RESEARCH

Two major manufacturers, AMD and Intel, share the market of processors. In the medium segment of performance Intel offered processors at about the same price as AMD In the lower segment of performance Intel's processors are somewhat more expensive than AMD's. The sources engaged in comparing performance of computer hardware argue that AMD offered the best price/performance ratio in early 2007, according to (PriceGraber, 2009).

[FIGURE 1 OMITTED]

It should be mentioned that the processors included in the sample that we analysed mainly belong to the category of smaller performance index of 2. So the conclusion is that in this sphere of approximately the same performance of the processors manufactured by AMD and Intel, Intel processor is generally more expensive. Depending on the class range, the price of processors in Croatia (2007) varied in that Intel's processors were more expensive from 10 to 50%.

When buying a new computer (the whole configuration), buyers have to decide on the best performance. Performances depend on all components of the computer (configuration), but are primarily determined by the central processor CPU. Based on that component, customers should choose all other components.

4. RESULTS

In view of the belief that Intel is somewhat more oriented to business users, whereas AMD is more inclined to home users and plays, it should be noted that in the total number of purchases, 2/3 of customers choose AMD and 1/3 Intel. The results of the AMD customer segment who bought their processors are shown in Table 1. Some of the results of Intel processors customers are shown in Table 2.

We have noted a great similarity of results because we have defined that the threshold for all values is 60%. Only with the threshold of 65% for the values of support and confidence would challenge the associative rule for the buyers of Intel processors and their decision to buy lower RAM memory. As for a buyer of AMD processor, it would increase to 73%. The rule that they will buy a 160Gb HDD (and higher) with the chosen threshold of 60% is satisfied for AMD customers. Intel's customers will definitely not buy a HDD with size less than 160Gb.

There is high probability that both categories of customers will not choose a processor in pair with a 17 "monitor, but it is also highly probable that the buyer of AMD or Intel processors will not buy a 19" or larger monitor at the same time. This is probably because of configuration upgrading--most of customers already have their old computer configurations, including monitors, or buy their monitors separately from others parts.

It should be noted that a significant number of customers did not have a monitor on the receipt bill with the processor and other essential components, and monitor is an integral part of the overall configuration. This may be a consequence of the fact that many of customers already have their own configurations with monitors included and only make an upgrade.

5. CONCLUSION AND FURTHER RESEARCH

The research in a typical Croatian retail unit that sells PC equipment has shown that the market segment (the number of customers) of AMD processors was approximately twice the size of the segment of Intel processors. This is probably because of the higher prices of Intel processors. But, customers who decide on more expensive Intel processors within their configurations are not particularly prone to buy more memory, bigger monitors, or more RAM.

This result is somewhat surprising, since the customer who pays more for an Intel processor is expected to buy other equipment of better quality. However, we also stress that the method of associative rules applied to our sample, showed that the probability threshold of choosing smaller RAM is lower with the Intel processors segment of customers.

The results of this research could give the support in configuring of computer systems for wide use, especially in first of three sub-processes of the product configuration process--in statement of construction (Brown 1998, Hansen et al., 2003).

Our further research will be aimed at making comparison between the association rules with respect to changes during period of time. For this purpose we intend to include more recent data in our application. Also, we plan to extend research methods and to treat customers behavior with decision three analysis and neural network methodology.

6. REFERENCES

Brown, D.C. (1998). Defining configuring, Artificial Intelligence for Engineering Design, Analysis and Manufacturing, Vol. 12, No. 4, pp 301-305

Hansen, T.; Scheer, C. & Loos, P. (2003). Product Configurators in Electronic Commerce--Extension of the Configurator Concept towards Customer Suggestions. In: Reichwald, R.; Piller, F.; Tseng, M. (Hrsg.): Proceedings of the 2nd Interdisciplinary World Congress on Mass Customization and Personalization (MCPC), Munchen

Loshin, D. (2003). Business Intelligence--The savvy manager's guide, Elsevier Science, USA, pp 205.

Maimon, O. & Rokach, L. (2005). The Data Mining and Knowledge Discovery Handbook. Springer

Moss, L. & Atre S. (2003). Business Intelligence Roadmap, Addison-Wesley, Boston

Mylonakis, J. (2007). A Research Study of Customer Preferences in the Home Loans Market: The Mortgage Experience of Greek Bank Customers, International Research Journal of Finance and Economics, ISSN 14502887 Issue 10 (2007)

Osterwalder, A. & Pigneur, Y. (2003). Modelling Customer Relationships in e-Business, 16th Bled Electronic Commerce Conference eTransformation, http:// www.online-baze.hr, Accessed on: 2009-09-08

Raby, M. (2009). Other AMD dual-cores follow price downswing, TG Daily, http://www.tgdaily.com/content/ view/31427/137/, Accessed on: 2009-07-03

Raby, M. (2009). Other AMD dual-cores follow price downswing, TG Daily, http://www.tgdaily.com/ content/view/31427/137/, Accessed on: 2009-09-09

Silverman, B. G., Bachann, M. & Al-Akharas, K. (2001). Implications of Buyer Decision Theory for Design of eCommerce Websites

*** (2009) http://computers.pricegrabber.com/ processorsretail-box/p/42/st=category/ --Pricegrabber, Accessed on: 2009-09-12

Tab. 1. Summary of association rules (AMD table). Min.
support = 60,0%, Min. confidence = 60,0%, Min. correlation =
60,0% Max. size of body = 10, Max. size of head = 10

Body Head Support Confidence Correlation
 (%) (%) (%)

AMD RAM<=512 MB 72,22 72,22 84,98
AMD not [RAM >=1 GB] 86,11 86,11 92,79
AMD HDD >=160 GB 61,11 61,11 78,17
AMD not [LCD 17"] 69,44 69,44 83,33
AMD not [LCD >=19"] 72,22 72,22 84,98

Tab. 2. Summary of association rules (Intel table). Min. support
=60,0%, Min. confidence = 60,0%, Min. correlation = 60,0%
Max. size of body = 10, Max. size of head = 10

Body Head Support Confidence Correlation
 (%)
 (%) (%)

Intel RAM<=512 MB 64,70 64,70 80,43
Intel not [RAM >=1 GB] 82,35 82,35 90,74
Intel not [HDD<160 GB] 70,58 70,58 84,01
Intel not [LCD 17"] 70,58 70,58 84,01