Integration of Machine Learning and Legal Requirements Analysis in a Mining Formalization Support System in Peru

Charca-Hidalgo H.J.; Gomez-Quispe H.Y.; Arizaca-Avalos A. · International Journal of Intelligent Engineering and Systems (2026) · DOI: 10.22266/ijies2026.0630.27

Resumen

To give a sense of the degree of informal status in the mining industry, artisanal and small-scale mining (ASM) in Peru employs over 80,000 operators resulting in tax losses, environmental hazards, and job insecurity. One bottleneck is the timely verification of the Enabling Requirements of the REINFO file to achieve the administrative step in the Comprehensive Mining Formalization Process to initiate/restart work. This paper proposes a hybrid decision-support framework focusing (i) on automatic verification of legal needs and (ii) predictive models of file prioritization. A national REINFO database had been designed and longitudinally linked to later formalization outcomes, thus controlling the temporality of predictors and outcome definition with less potential risk of information leakage. Using a modified longitudinal design, the analyzed cohort consisted of 87,300 baseline REINFO data points with 279 being associated with events after baseline formalization, with a corresponding prevalence of 0.3196. A chronological evaluation scheme was adopted for evaluation and supervised models were evaluated with respect for rank-based criteria due to this being a rare-event outcome. The extended specification achieved ROC-AUC = 0.9779 and PR-AUC = 0.3096 in the temporal test set. In terms of operation, 35 of 56 positives were included in an optimal set within the top 1% of ranked files (precision@1% = 0.2000, recall@1% = 0.6250, lift = 62.3571), while 46 were included in the top 5% (recall@5% = 0.8214, lift = 16.4286). It shows that useful prioritization is still possible under tight temporal definition of the milestone. The two most popular predictive factors overlap with the hierarchical design of the implementation, in particular its requirement to provide processing with mining rights, technical files, surface land accreditation, and CIRA. The architecture offers it by using a rules engine for deterministic compliance checking together with a machine learning module for estimating a probabilistic ranking score, and the traceability and explainability are still maintained. Such hybrid model provides a technically sound basis for future use in PFMI/REINFO management, however, calibration and territorial robustness and fairness evaluation remain to be performed along with prospective validation under actual administrative conditions. This article is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License. License details: https://creativecommons.org/licenses/by-sa/4.0/

Instituciones: Universidad Nacional del Altiplano