Quality of Artificial Intelligence Driven Procurement Decision Making and Transactional Data Structure

Radoslav Delina, Marek Macik

Quality of Artificial Intelligence Driven Procurement Decision Making and Transactional Data Structure

Číslo: 1/2023
Periodikum: Quality Innovation Prosperity
DOI: 10.12776/qip.v27i1.1819

Klíčová slova: transactional data; public procurement; prediction; data structure; machine learning

Pro získání musíte mít účet v Citace PRO.

Přečíst po přihlášení

Anotace: Purpose: Current data driven decision making development calls for the quality assurance based on quality data structure. The paper analyses transactional data structure used in public procurement in Slovakia and the effect of data structure enhancement on prediction performance as crucial part of artificial intelligence (AI) quality assurance standard. We examine the significance of data structure enhancement and attributes transformation for prediction modelling.

Methodology/Approach: The research is based on mutli-step model using stacked ensemble machine learning (ML) algorithm and simulating input space of 211 attributes transformed and aggregated according to different perspectives assessed by r2, mean absolute error (MAE) or mean square error (MSE).
Findings: The results show that different performance of variable categories to prediction power. The most significant predictors were in category related to sectoral product classifications and in category related to variables aggregated for supplier, what underline the significance of structured information of all suppliers and negotiation participants in public tenders.
Research Limitation/Implication: Methodology is based on big data with high complexity. Due to limited computing power, no subjects’ IDs were used as inputs. The complexity behind data and processes call for more complex simulations of all variables and their mutual interaction and interdependencies.
Originality/Value of paper: The paper contributes to data science in transactional data domain and assessed the significance of different variables categories with respect to their specific added value to prediction power.