Derivation and test of spatial rules for the prediction of efficiency of business branches
Ara Toomanian · 2007
In recent years, optimal site selection has become one of the main concerns for managers of business enterprises. In addition, various kinds of spatial and non-spatial parameters influence the efficiency of new branches. These factors have a direct relation with site selection indicators. In this research project, we use data mining algorithms to find and extract useful knowledge not only to help managers making better decisions for site selection, but also for extracting associations between parameters at different scales. We also tried to find a link between a mathematically determined efficiency measure and spatial association rules, which is a database method in data mining. During the research, we classified the study area into three different classes as ‘high’, ‘average’, and ‘low’ according to the efficiency and turnover measures. Afterwards, in each class we used an a priori like algorithm to find the most frequent item sets and predict an average range of efficiency. For the second scenario we also put a limitation in the a priori like algorithm to derive a constraint-based frequent item set containing the efficiency measure parameters. In general, as the efficiency measure in the low class had a higher frequency than in other classes, we obtained negative rules rather than positive rules. In addition, the association rules for the small scale gave more meaningful results than those of the large scale. The reason was in the use of real parameters instead of aggregated parameters. The usability of this method was not absolutely good with this data set and we recommend to use normal distributed efficiency measure data to find association rules in all the classes. Finally, for the site selection issues, the managers can use this method as a comparison factor, among different candidate areas. They can rely on the validation measures such as support, confidence, lift and leverage to select the best location for a new site. Needless to say, such models can be used for the parameters inside the frequent item set.