AN INTEGRATED AHP–TOPSIS–KNAPSACK–LLM HYBRID MODEL FOR PROJECT PORTFOLIO OPTIMIZATION
Abstract
This paper proposes an integrated four-component hybrid pipeline for project portfolio optimization that combines the Analytic Hierarchy Process (AHP) for criterion weighting, TOPSIS for alternative ranking, 0/1 Knapsack dynamic programming for budget-constrained optimal selection, and GPT-4o based large-language-model (LLM) scoring for quantifying qualitative project documents. The mathematical apparatus, overall computational complexity O(n·m + n·B), and Design Science Research based experimental protocol are formalized. Validation is planned on a synthetic dataset of 20 projects across 30 replications against three baselines (pure NPV greedy, greedy heuristic, and pure MCDM). Four falsifiable hypotheses (H1–H4) are tested using paired t-tests, the Wilcoxon signed-rank test, and Monte Carlo sensitivity analysis. The novelty of the work lies in being the first fully integrated AHP+TOPSIS+0/1 Knapsack+LLM pipeline reported in the literature, with direct applicability to the IT initiatives planned under the "Digital Uzbekistan 2030" strategy.