Page 56 - ISC PROCEEDINGS 21.4

medium-term public investment plans, annual ST estimates and allocations, real-time
disbursement data through the State Treasury system, e-procurement records,
construction contracts, final settlement reports, as well as audit and inspection findings.
The fragmentation and structural inconsistency of these datasets reduce integration and
analytical capacity in the absence of an appropriate governance framework.
The Big Data Governance framework in this study comprises four main components:
First, data standardization and interoperability. Establishing unified identification
codes (e.g., project codes, investor codes, contract codes, and ST classification codes) is a
prerequisite for integrating data across Public financial governance systems.
Standardization enables lifecycle tracking of projects from planning to final settlement.
Second, data quality assurance. Evaluation criteria include completeness, accuracy,
timeliness, and consistency. In an AI-driven environment, poor data quality may lead to
model bias and unreliable forecasts.
Third, data security and information protection. PIE data involves ST, contracts, and
enterprises; therefore, access control mechanisms, encryption, and cybersecurity
safeguards are required, consistent with public-sector data protection standards.
Fourth, data responsibility and ethics. Data use must adhere to principles of
transparency, avoid misuse of information, and ensure legality in automated processing.
Big Data Governance serves as the institutional and technical foundation for
deploying advanced analytical tools in the monitoring of public investment.
(ii) Predictive Analytics
Predictive Analytics refers to the use of historical data combined with statistical and
ML models to forecast future trends and risks. In PIE management, this approach enables
a shift from reactive control to proactive risk management.
Theoretically, predictive analytics is based on the assumption that patterns
observed in historical data can indicate the probability of future events. For instance, data
on projects that experienced delays, cost adjustments, or design changes can be used to
build risk-scoring models for new projects.
The techniques examined in this study include multivariate regression, decision
trees, random forest models, and deep learning approaches. Applications in public
investment oversight may include: Forecasting the likelihood of disbursement delays
based on actual progress and investor capacity; Detecting anomalies in payment flows
relative to planned capital allocations; Identifying factors that increase the risk of cost
overruns; Ranking inspection priorities for projects with high-risk scores.
However, applying predictive analytics in the public sector requires ensuring
algorithmic transparency and explainability to meet accountability standards.
(iii) AI for Public Accountability
Accountability is a core principle of public financial governance, requiring state
agencies to be transparent in ST utilization and subject to oversight by relevant
stakeholders. AI for Public Accountability refers to the application of AI technologies to
enhance transparency and strengthen checks on public authority.
This theoretical framework is structured along three dimensions:
First, enhancing transparency. AI can automate ST data analysis and visualization,
supporting real-time information disclosure. Natural Language Processing (NLP)
technologies can review contracts and legal documents to detect unusual clauses.
Second, strengthening oversight capacity. AI systems function as early-warning
mechanisms, detecting anomalous transactions in payments or procurement processes.

55