Use of data mining techniques to create reference pricing approach for the Rwanda Public procurement

Abstract

In order to support e-procurement operation, Rwanda Public Procurement Authority (RPPA) initiated and implemented policies of using reference prices as tool to strengthen public procurement practice. The reference prices are used by procuring entities (public institutions) in evaluation of tenders. Currently, reference price is calculated manually by gathering information from different suppliers on the market place located in Kigali and other districts of Rwanda. Due to the manual process, the update of the previous reference price long time. The main purpose of this study was to develop a digitalized approach of creating reference prices by using data from online e-procurement system for Rwanda. Additionally, the study purpose of this study was to show how k-means clustering with text vectorization method used to create a reliable reference price for similar items. Text clustering was applied to identify groups of similar services or products. Text vectorization methods, namely Bag of words and tf-idfVectoriser are investigated in clustering similar item using k-means clustering. Tf-idfVectoriser fitted k-means clustering method with robustness and optimal clusters of 13 obtained from elbow method. Thus, tf-idfVectoriser was the best method in creation of reference price per similar item. Finally the average reference price per item were computed, the results proved that there exist large standard deviation between prices of similar services or products. This deviation resulted in abnormally higher reference price per item. To avoid the abnormality, the median was measured for the purpose of comparison with the obtained average price. In this case, the median prices per similar item performed well than the average price based on the existing market price plus other related costs. Therefore, for products and services category of items, the study accepted the median prices as the reliable reference prices per item.