The purpose of this article is to conduct a systematic literature review on the role of deep learning in credit risk prediction for fintech lending, with particular emphasis on model architectures, imbalanced data handling techniques, and the mathematical foundations underpinning these methods. Open-access scientific publications retrieved from three complementary databases (Scopus, IEEE Xplore Digital Library, and Web of Science Core Collection) were used to conduct the systematic literature review. Following the PRISMA 2020 protocol, 30 publications were selected after a rigorous multi-stage screening process involving deduplication across databases, temporal filtering (2015–2026), and thematic eligibility assessment. Data for analysis were processed using Python-based bibliometric tools and network analysis (replicating R Bibliometrix and VOSviewer functionalities). The results of the analysis indicate a sustained growth in research on deep learning applications for fintech credit risk, with a compound annual growth rate (CAGR) of approximately 29.2% and an average of 27.17 citations per document. The segmentation of the studied conceptual landscape made it possible to identify four interconnected thematic clusters: (1) peer-to-peer lending and default prediction architectures; (2) explainability and XAI-based methods for credit scoring; (3) imbalanced data and hybrid deep-learning frameworks; and (4) credit risk assessment combining deep learning and statistical approaches. The following research areas on the deep learning-based transformation of fintech credit risk prediction have been identified: (1) feedforward deep neural networks and attention-based architectures (LSTM, CNN) as dominant predictive engines; (2) hybrid deep ensemble and deep-boosting frameworks (e.g., LightGBM-Attention, GBDT-Deep FFM) as emerging high-performance paradigms; (3) specialized techniques for imbalanced data handling—including ADASYN, SMOTE, cost-sensitive learning, and balanced stratified prioritized experience replay—as critical methodological frontiers; and (4) transformer-based and XAI-integrated architectures as the emerging frontier. The originality of this article lies in its explicit focus on the mathematical and methodological challenges of deep learning-based credit risk prediction in fintech lending, providing an actionable research agenda that addresses class imbalance, uncertainty quantification, loss function design, concept drift, and regulatory compliance. The findings provide valuable insights for scholars, practitioners, and policymakers, and outline a concrete roadmap for developing more accurate, robust, and explainable credit risk models in the rapidly evolving fintech ecosystem.

Cardinality-constrained portfolio optimization becomes computationally difficult when downside-risk control, minimum holdings, transaction costs, and diversification requirements are considered simultaneously. Existing clustering-assisted approaches may preselect cluster representatives, impose rigid cluster quotas, or use scale-sensitive concentration penalties, potentially discarding attractive assets, over-restricting the feasible region, or providing no explicit control over deterioration of the primary risk objective. This study proposes ε-CAGWO–LP (epsilon-lexicographic Cluster-Aware Grey Wolf Optimizer–Linear Programming), a matheuristic for cardinality-constrained Mean–Conditional Value at Risk (Mean–CVaR) portfolio selection. The method separates discrete asset selection from continuous allocation: an adaptive Grey Wolf Optimizer (GWO) explores fixed-cardinality subsets, whilelinear programming (LP) determines the optimal weights, historical CVaR, and turnover for each evaluated subset. Assets are dynamically grouped using K-Medoids and a hybrid distance combining downside correlation with robust risk–return characteristics. Phase 2 reduces cluster concentration only among portfolios whose primary Mean–CVaR objective remains within a prespecified tolerance relative to the internal Phase 1 reference. The framework was evaluated on 45 Indonesian equities observed from 2018 to 2025 using 756-day rolling training windows and 60 monthly out-of-sample rebalances. Relative to Adaptive Grey Wolf Optimizer–Linear Programming (AGWO–LP), ε-CAGWO–LP reduced mean cluster Herfindahl–Hirschman Index (HHI) from 0.368 to 0.338 (8.24%) and increased mean cluster coverage from 3.93 to 4.10. All final portfolios satisfied the 0.25% epsilon limit relative to the internal Phase 1 reference, with a maximum positive objective deterioration of 0.239%. The method recorded a cumulative return of 93.75%, an annualized return of 14.83%, and a Sharpe ratio of 0.877, but its return advantage over AGWO–LP and equal weighting was not statistically significant. Exact-solution experiments showed that mixed-integer linear programming (MILP) remained preferable for approximately 40 assets, whereas the matheuristic became more computationally attractive for larger universes. From a broader development perspective, the proposed risk-aware portfolio decision-support framework is relevant to the economic-growth dimension of Sustainable Development Goal 8 (SDG 8), particularly in the context of investment and capital-allocation decisions. Overall, the principal empirical benefit was controlled structural diversification within the prescribed primary-objective tolerance rather than demonstrated return superiority.

This study aims to identify and analyze the application of the Mean-Conditional Value-at-Risk (Mean-CVaR) model in the allocation of financial asset portfolio weights combined with the K-Means Clustering algorithm. The Systematic Literature Review (SLR) method is used with the PRISMA protocol through the stages of identification, screening, eligibility, and inclusion. Data is obtained from Scopus, ScienceDirect, and Dimensions databases, then selected up to six relevant primary articles. The results of the study indicate that CVaR is the dominant risk measure in portfolio optimization, while K-Means Clustering serves as a method of grouping assets to increase diversification. The optimization methods used include Genetic Algorithm, Particle Swarm Optimization, Teaching Learning-Based Optimization, and Stochastic Programming. However, direct integration between Mean-CVaR and K-Means within a portfolio weight allocation framework is still rare. This research emphasizes the need to develop a hybrid model that combines both approaches in an integrated manner, applied to a multi-asset portfolio, and validated under various market conditions to produce an optimal, adaptive, and resilient investment strategy against extreme risks.

Economic evaluation of Battery Energy Storage Systems (BESSs) integrated with solar photovoltaic (PV) systems requires the consideration of financial performance, operational behavior, battery degradation, reliability, and uncertainty. This PRISMA-based systematic review synthesizes mathematical and decision-making approaches relevant to Fuzzy Soft Set-based PV–BESS evaluation. Searches in Scopus, ScienceDirect, SpringerLink, and Taylor & Francis Online identified 789 records published between 2015 and 2026, and 74 studies were included. All studies were coded by theme, method, and economic and technical criteria, while 15 representative studies were analyzed at the equation level. Six model families were identified: techno-economic valuation, multi-objective planning, operational energy management, reliability assessment, fuzzy and multi-criteria decision-making, and degradation-aware optimization or control. Most studies appeared in 2024–2026 (57/74; 77.0%), while keyword analysis highlighted decision making and battery storage. Common formulations included NPV, NPC, LCOE, IRR, payback period, market revenue, battery capacity fade, equivalent full cycles, degradation cost, LPSP, LOLE, EENS, Monte Carlo simulation, and fuzzy ranking methods. These components remain fragmented across separate frameworks. Fuzzy Soft Set is therefore positioned as a potential parameterized decision-support layer for integrating heterogeneous outputs, subject to further theoretical development and empirical validation. This direction is relevant to Sustainable Development Goal 7 (Affordable and Clean Energy).

Rainfall-index-based disaster insurance is an efficient approach to mitigating hydrometeorological losses. However, conventional premium pricing models generally assume memoryless stochastic dynamics that do not fully capture the long-range dependence inherent in rainfall data. This study develops a hydrometeorological disaster insurance model within a fractional Black–Scholes framework to incorporate long-memory effects. The model is formulated using fractional differential equations and solved semi-analytically by integrating the Daftardar–Jafari Method (DJM) with the Kashuri–Fundo (KF) transform, yielding a closed-form solution expressed in terms of the Mittag–Leffler function. The proposed contract is structured as parametric rainfall insurance with a multi-layer payout mechanism based on percentiles corresponding to minor, moderate, and severe housing damage. The results show that variations in the fractional-order parameter significantly affect premium estimation. In particular, $\delta =0.5$ recovers the classical model and tends to generate higher premiums than the fractional model with ${\delta }_1=0.23153$, whereas the model with ${\delta }_2=0.73153$ yields lower premiums. These findings indicate that fractional-order parameterization can accommodate diverse risk characteristics and policyholders’ economic capacities, enabling more adaptive, risk-sensitive premium structures. In line with SDG 13 (Climate Action), the proposed framework offers a climate-responsive disaster-mitigation strategy through accessible, actuarially relevant insurance design.