RISK PREDICTION MODEL IN LOGISTICS MANAGEMENT USING ARTIFICIAL INTELLIGENCE AND DIGITAL PLATFORMS

Jalolova Ruxshona Nosir qizi

Ubaydullayeva Fariza Sheraliyevna

O’rinboyeva Zarina Xabibullo qizi

Samarqand Institute of Economics and Service,

Researchers

Annotatsiya. Ushbu maqolada logistika menejmentida sun’iy intellekt va raqamli platformalar yordamida risklarni bashorat qilish modelini yaratish va qo’llashning ilmiy-uslubiy asoslari ko’rib chiqilgan. Tadqiqot natijalari shuni ko’rsatadiki, mashinali o’qitish algoritmlari va real vaqtdagi ma’lumotlar tahlili bilan jihozlangan raqamli platformalar logistik risklarni aniqlashning aniqligini sezilarli darajada oshiradi va zanjir bo’ylab ta’minot samaradorligini yaxshilaydi. O’zbekiston korxonalari va logistik tashkilotlari uchun sun’iy intellektga asoslangan risk boshqaruv tizimini joriy etishga doir amaliy tavsiyalar ishlab chiqilgan.

Kalit so’zlar: sun’iy intellekt, logistika menejmenti, risk bashorat qilish, raqamli platformalar, mashinali o’qitish, ta’minot zanjiri, real vaqtdagi monitoring, prognoz tahlili.

Аннотация. В данной статье рассматриваются научно-методологические основы разработки и применения модели прогнозирования рисков в управлении логистикой с использованием искусственного интеллекта и цифровых платформ. Результаты исследования показывают, что цифровые платформы, оснащённые алгоритмами машинного обучения и анализом данных в режиме реального времени, значительно повышают точность выявления логистических рисков и улучшают эффективность цепочки поставок. Разработаны практические рекомендации по внедрению системы управления рисками на основе искусственного интеллекта для предприятий и логистических организаций Узбекистана.

Ключевые слова: искусственный интеллект, управление логистикой, прогнозирование рисков, цифровые платформы, машинное обучение, цепочка поставок, мониторинг в реальном времени, прогностический анализ.

Abstract. This article examines the scientific and methodological foundations for developing and applying a risk prediction model in logistics management using artificial intelligence and digital platforms. Research findings demonstrate that digital platforms equipped with machine learning algorithms and real-time data analytics significantly enhance the accuracy of logistics risk identification and improve supply chain efficiency. Practical recommendations are developed for implementing an AI-driven risk management system for enterprises and logistics organizations in Uzbekistan.

Key words: artificial intelligence, logistics management, risk prediction, digital platforms, machine learning, supply chain, real-time monitoring, predictive analytics.

Introduction

The rapid expansion of global trade networks and the increasing complexity of supply chains have made logistics risk management one of the most critical challenges for modern enterprises. Disruptions caused by geopolitical shifts, pandemic-driven demand volatility, transportation bottlenecks, and fluctuating fuel prices have exposed the vulnerability of traditional reactive risk management approaches. In this context, the integration of artificial intelligence (AI) and digital platforms into logistics operations has emerged as a transformative solution for proactive risk identification and mitigation.

Artificial intelligence, particularly machine learning (ML) and predictive analytics, enables logistics managers to process vast volumes of structured and unstructured data from multiple sources — including market signals, weather forecasts, supplier performance records, and historical shipment data — to generate actionable risk predictions in near real time. Digital platforms serve as the connective infrastructure that aggregates these data streams, applies analytical models, and delivers decision-support outputs to stakeholders across the logistics network.

Uzbekistan’s logistics sector is undergoing significant transformation as the country positions itself as a regional transit hub along the reconstructed Silk Road trade corridors connecting China, Central Asia, and Europe. Presidential Decree No. PF-60 (2022) on the development of logistics infrastructure and the “Digital Uzbekistan 2030” strategy explicitly prioritize the digitalization of transport and logistics operations. Despite these policy commitments, the adoption of AI-driven risk prediction tools among Uzbek logistics enterprises remains nascent, with most companies still relying on manual reporting and experience-based judgment.

The purpose of this article is to propose a structured risk prediction model for logistics management that leverages AI and digital platforms, evaluate its effectiveness through empirical research conducted at enterprises in the Samarqand and Tashkent regions, and formulate implementation recommendations adapted to the Uzbek business environment.

Literature Review and Research Methodology

The theoretical foundations of AI-driven logistics risk management draw upon several interconnected academic streams. Christopher (2016) established the conceptual framework for supply chain risk management, categorizing logistics risks into supply-side, demand-side, and environmental disruptions. His taxonomy remains foundational for contemporary AI model design, as it defines the scope of variables that predictive algorithms must account for.

The application of machine learning to supply chain risk prediction was systematically analyzed by Nguyen et al. (2018), who demonstrated that ensemble learning methods — particularly Random Forests and Gradient Boosting — outperform traditional statistical models in predicting delivery delays and supplier defaults. Their findings highlighted the critical importance of feature engineering: selecting and transforming raw logistics data into meaningful input variables for ML models.

Ivanov and Dolgui (2020) introduced the concept of the “ripple effect” in supply chains — the propagation of localized disruptions across interconnected logistics networks — and argued that AI-based digital twins represent the most effective tool for modeling and mitigating such cascading risks. Their simulation studies showed that AI-enhanced digital twins can reduce recovery time from supply chain disruptions by up to 30 percent compared to conventional contingency planning.

Within the Central Asian academic context, Nazarov (2021) examined the readiness of Uzbek enterprises to adopt digital logistics solutions, identifying infrastructure gaps and human capital shortages as the principal barriers. Tursunova (2022) analyzed the role of the Uzbek Electronic Logistics Platform (ELP) in improving freight transparency and argued for the integration of predictive analytics modules into existing digital infrastructure. Karimov (2023) proposed a preliminary framework for AI-based risk assessment in Uzbek transit logistics, though his model was not empirically tested.

The research methodology of this study combines quantitative and qualitative approaches. Structured surveys were administered to 45 logistics managers and supply chain professionals at enterprises in Samarqand, Tashkent, and Bukhara regions during 2023–2024. Additionally, operational data from two pilot implementations of the proposed risk prediction model at logistics companies were collected and analyzed. Statistical analysis was performed using descriptive statistics and comparative performance metrics, including precision, recall, and F1-score for model evaluation.

Analysis and Discussion of Results

Architecture of the Proposed Risk Prediction Model

The proposed model operates across four integrated layers: data acquisition, preprocessing, predictive modeling, and decision-support output. In the data acquisition layer, the digital platform aggregates inputs from IoT sensors embedded in transport vehicles, ERP systems, external market data APIs, weather services, customs databases, and supplier performance records. This multi-source architecture ensures that the model captures both internal operational variables and external risk drivers.

The preprocessing layer applies data cleaning, normalization, and feature extraction routines. Given the heterogeneity of logistics data — combining numerical, categorical, and temporal variables — the platform employs automated machine learning (AutoML) pipelines that adapt preprocessing steps to incoming data characteristics. Missing values, particularly common in supplier reporting, are addressed through k-nearest-neighbor imputation rather than simple mean substitution, preserving distributional properties.

The predictive modeling layer houses a hybrid ensemble model combining three base learners: a Gradient Boosting Machine for structured tabular data, a Long Short-Term Memory (LSTM) recurrent neural network for time-series forecasting of demand and transit delays, and a Bayesian Network for probabilistic reasoning under uncertainty. The ensemble integrates outputs via a stacking meta-learner that weights each base model’s predictions according to historical accuracy on validation data.

The decision-support output layer translates model predictions into risk scores categorized along two dimensions: probability (low, medium, high) and impact severity (minor, moderate, critical). Risk scores are visualized through the platform’s dashboard interface and trigger automated alert protocols when thresholds are exceeded, enabling supply chain managers to initiate contingency responses without delay.

Table 1

Risk Categories and AI Model Components

Risk Category	Primary Data Sources	AI Model Applied
Supply disruption	Supplier KPIs, procurement records	Gradient Boosting + Bayesian Net
Transit delays	GPS telemetry, weather APIs, customs data	LSTM Neural Network
Demand volatility	ERP sales data, market signals	LSTM + Gradient Boosting Ensemble
Warehouse capacity	WMS data, IoT sensors	Regression + Rule Engine
Regulatory/compliance	Customs databases, policy updates	Bayesian Network

Source: Compiled by the authors

Empirical Research Results

The risk prediction model was piloted at two logistics enterprises in the Samarqand region over a six-month operational period (March–August 2024). Enterprise A specializes in agricultural commodity transport, while Enterprise B operates a third-party logistics (3PL) service handling manufactured goods. Both enterprises maintained a control period using their legacy risk management systems before switching to the AI-powered platform for the pilot phase.

Performance was measured across five key indicators: risk prediction accuracy, lead time for risk identification, frequency of unplanned disruptions, inventory holding costs associated with buffer stock maintained against risk uncertainty, and overall supply chain resilience score as rated by enterprise managers. The comparative results are presented in Table 2.

Table 2

Performance Comparison: AI Risk Platform vs. Traditional Risk Management

Performance Indicator	AI Platform (%/score)	Traditional System (%/score)
Risk prediction accuracy	84%	51%
Avg. lead time for risk ID (days)	2.1 days	7.4 days
Unplanned disruption frequency	Reduced by 61%	Baseline
Inventory buffer cost reduction	23% savings	Baseline
Manager-rated resilience score	8.2 / 10	5.1 / 10
On-time delivery rate	91%	73%

Source: Compiled by the authors based on pilot study data (Samarqand region, 2024)

The results demonstrate that the AI-powered risk prediction platform achieved a 33-percentage-point improvement in risk prediction accuracy compared to the traditional system. The reduction in average lead time for risk identification — from 7.4 days to 2.1 days — is particularly significant, as earlier risk detection allows for longer response windows and lower disruption costs. The 61 percent reduction in unplanned disruptions and the 23 percent reduction in inventory buffer costs represent direct economic benefits for enterprise operations.

Discussion

Strengths of the AI-Driven Approach

The most fundamental advantage of AI-driven risk prediction in logistics lies in its capacity for pattern recognition across high-dimensional datasets that exceed human cognitive processing capacity. A logistics manager reviewing supplier performance reports, weather forecasts, and market signals simultaneously faces cognitive overload; an ML model can process these data streams continuously and integrate them into coherent risk signals. This scalability is particularly valuable as Uzbekistan’s trade volumes grow and supply chains become more complex.

A second major advantage is the model’s ability to learn and adapt over time. Unlike static rule-based systems that require manual updates when operational contexts change, the ensemble model continuously retrains on new data, improving its accuracy as the enterprise accumulates more operational history. Survey respondents at the pilot enterprises rated the self-learning capability as the most valued feature, with 89 percent indicating it reduced the burden of manual risk monitoring on their teams.

The digital platform infrastructure also enables unprecedented transparency in risk communication across the supply chain. Suppliers, carriers, warehousing partners, and customers can access risk dashboards relevant to their role, enabling collaborative risk mitigation rather than siloed decision-making. This network-level transparency is aligned with the principles of integrated supply chain management advocated by Chopra and Meindl (2021).

Limitations and Implementation Challenges

Despite the strong empirical results, several implementation challenges were identified during the pilot study. Data quality emerged as the most significant barrier: at both pilot enterprises, a substantial proportion of supplier reporting was incomplete or inconsistently formatted, requiring extensive preprocessing effort before the model could generate reliable predictions. This finding underscores that AI model performance is fundamentally contingent on the quality and completeness of input data.

The initial implementation costs also present a barrier, particularly for small and medium-sized logistics enterprises (SMEs) that constitute the majority of Uzbekistan’s logistics sector. Hardware infrastructure for IoT sensor networks, licensing fees for cloud-based ML platforms, and the cost of integrating the AI system with existing ERP and WMS software require capital investments that many SMEs cannot readily absorb. Subsidized access through government-backed digitalization programs or shared-infrastructure models may be necessary to democratize access.

Additionally, the shortage of data science and AI engineering talent in Uzbekistan’s logistics sector represents a human capital constraint. Survey results indicated that only 18 percent of logistics managers surveyed reported having sufficient in-house expertise to maintain and interpret AI-driven systems. Without ongoing technical support, there is a risk that platforms degrade in performance as data environments evolve and retraining is neglected.

Opportunities and Recommendations for Uzbekistan

The “Digital Uzbekistan 2030” strategy and the ongoing development of the Uzbekistan Logistics Center (ULC) as a regional hub provide a strategic foundation for accelerating AI adoption in logistics risk management. The government’s existing investment in digital infrastructure, including the expansion of fiber-optic networks to Samarqand, Bukhara, and Namangan regions, reduces the connectivity barriers that impede platform deployment in peripheral areas.

Several concrete measures are recommended to translate strategic policy into operational implementation. First, the Ministry of Transport and the Agency for the Development of the Digital Economy should jointly establish a Logistics AI Sandbox — a publicly accessible environment where enterprises can pilot AI risk management tools with subsidized access to cloud computing resources and expert technical support. This model has proven effective in Singapore and Kazakhstan and is well-suited to Uzbekistan’s development context.

Second, collaboration between universities — particularly Samarqand Institute of Economics and Service, Tashkent State Technical University, and Westminster International University in Tashkent — and logistics enterprises should be formalized through joint research programs focused on developing AI models calibrated to Central Asian logistics conditions. Uzbekistan’s transit corridor positioning creates unique risk dynamics (border crossing variability, multimodal handoff complexity) that global models may not adequately capture.

Conclusion and Recommendations

This research confirms that AI-driven risk prediction models integrated within digital platforms deliver measurably superior logistics risk management outcomes compared to traditional approaches. The pilot study demonstrated an 84 percent risk prediction accuracy rate, a 61 percent reduction in unplanned disruptions, and a 23 percent decrease in inventory buffer costs — outcomes that translate directly into competitive advantage and operational resilience for logistics enterprises.

The principal contribution of AI to logistics risk management lies not merely in automation but in the qualitative transformation of how organizations perceive and respond to uncertainty. By converting raw, heterogeneous operational data into probabilistic risk signals, AI enables managers to shift from reactive crisis response to proactive risk governance — a fundamental shift in organizational capability that becomes increasingly valuable as supply chain complexity grows.

Based on the research findings, the following recommendations are offered. First, the government should establish a Logistics AI Sandbox to provide SMEs with subsidized access to AI risk management tools and technical expertise. Second, logistics enterprises should invest in data governance frameworks to ensure the quality, completeness, and standardization of operational data — the prerequisite for effective AI model performance. Third, universities and research institutions should develop AI and data science curricula aligned with logistics sector needs to address the human capital shortage. Fourth, a national logistics risk data-sharing consortium should be established, allowing enterprises to pool anonymized operational data and collectively improve model accuracy. Fifth, international experience — particularly from Singapore, the Netherlands, and Kazakhstan — should be systematically studied and adapted to the Uzbek context through government-sponsored benchmarking programs.

References

1. Christopher, M. (2016). Logistics and Supply Chain Management (5th ed.). Pearson Education.

2. Nguyen, T., Zhou, L., Spiegler, V., Ieromonachou, P., & Lin, Y. (2018). Big Data Analytics in Supply Chain Management: A State-of-the-Art Literature Review. Computers & Operations Research, 98, 254–264.

3. Ivanov, D., & Dolgui, A. (2020). Viability of Intertwined Supply Networks: Extending the Supply Chain Resilience Angles towards Survivability. International Journal of Production Research, 58(10), 2904–2915.

4. Chopra, S., & Meindl, P. (2021). Supply Chain Management: Strategy, Planning, and Operation (7th ed.). Pearson.

5. Nazarov, B. (2021). Digital Readiness of Uzbek Enterprises in the Logistics Sector: Barriers and Enablers. Economics and Innovative Technologies, 3(2), 88–101.

6. Tursunova, G. (2022). The Uzbek Electronic Logistics Platform and Prospects for Predictive Analytics Integration. Transport and Communications, 4(1), 55–67.

7. Karimov, F. (2023). A Framework for AI-Based Risk Assessment in Uzbek Transit Logistics. Journal of Management and Digital Economy, 2(3), 34–49.

8. Decree of the President of the Republic of Uzbekistan No. PF-60. (2022). On Measures for the Development of Logistics Infrastructure. Tashkent.

9. Agency for the Development of the Digital Economy. (2023). Digital Uzbekistan 2030: Progress Report. Tashkent: Ministry of Digital Technologies.

10. World Bank. (2023). Logistics Performance Index: Uzbekistan Country Profile. Washington D.C.: World Bank Group.