Intelligent ELECTRE III Multi-Attribute Decision Making with Neural Network Threshold Detection and Multiprocessing Parallel Computation
Main article
Abstract
Multi-attribute decision-making (MADM) methods are fundamental tools for complex industrial and organizational decision problems, yet classical approaches face critical limitations when confronted with the scale and complexity demands of contemporary AI-driven environments. The ELECTRE III method, renowned for its nuanced treatment of preference heterogeneity through indifference, preference, and veto thresholds, suffers particularly from two structural challenges: the O(n^2) computational complexity of pairwise comparison that becomes prohibitive for large alternative sets, and the persistent difficulty of determining appropriate threshold parameter values for real-world applications. This paper proposes an intelligent ELECTRE III framework that addresses both limitations through integrated neural network and multiprocessing innovations. A multi-layer perceptron neural network is trained to automatically detect the three ELECTRE III threshold parameters (q, p, v) from criteria weight distributions and historical decision data, eliminating the expert elicitation bottleneck. A multiprocessing-based parallel ELECTRE III engine partitions the pairwise concordance and discordance matrix computations across available CPU cores, achieving near-linear speedup scaling. Evaluation on the QS World University Rankings dataset (n = 500 universities, 6 criteria) demonstrates that the neural network threshold detector achieves mean absolute errors of 0.011--0.021 across the three threshold types, while 8-core multiprocessing achieves 6.5× speedup at 1,000 alternatives. Comparative analysis against TOPSIS, VIKOR, PROMETHEE, AHP, and SAW confirms ranking consistency (Spearman rho > 0.88) while uniquely preserving the veto and incomparability structures that distinguish ELECTRE III from compensatory MADM methods. The proposed framework provides a practical pathway for deploying intelligent ELECTRE III in large-scale industrial supplier evaluation, technology selection, and strategic investment prioritization contexts.
