INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025
Page 1307
Intelligent Data Analytics Methods in Adaptive Inventory Planning
Kitaeva Iuliia
Master’s Degree, Financial University under the Government of the Russian Federation
DOI:
https://doi.org/10.51244/IJRSI.2025.1210000115
Received: 07 October 2025; Accepted: 15 October 2025; Published: 06 November 2025
ABSTRACT
This article examines intelligent data analytics methods applied in adaptive inventory planning under dynamic
market conditions. The possibilities of integrating demand forecasting with replenishment optimization models
aimed at improving the efficiency of supply chain management are analyzed. The role of machine learning
methods – including stochastic modeling, evolutionary algorithms, and reinforcement learning – in shaping
adaptive procurement and resource allocation strategies is examined. Particular attention is paid to minimizing
total costs while maintaining the required service level and increasing the resilience of logistics systems under
uncertainty.
Keywords – adaptive inventory planning; demand forecasting; machine learning; reinforcement learning;
stochastic modeling.
INTRODUCTION
Modern-day supply chains operate within surroundings of fluctuating consumer demand, geopolitical
turbulence, logistics upsets, and price movement in raw materials, and this requires the reevaluation of traditional
inventory management policies. Organizations are faced with the trade-off of balancing product availability and
cost reduction, which is becoming an increasingly difficult issue in a changing market environment.
Classical planning methods based on deterministic models and historical data have a certain stability, but they
do not reflect the full picture. Under conditions of uncertain demand or rapid fluctuations under the force of
external factors, static models are prone to either building up too much or, conversely, too little inventory. This
reduces the level of customer service and increases operational risks.
The availability of analytical technologies creates the possibility to leverage more responsive tools. Predictive
analytics and machine learning allow to uncover hidden patterns in big data and consider a wide range of factors
− from seasonal patterns to customer behavior via digital channels. These methods pave the way for the evolution
to adaptive management systems, where procurement and replenishment decisions are dynamically made with
forecasts updated continually.
In this regard, the relevance of the study lies in identifying the potential of intelligent data analytics methods for
transforming inventory management. The aim of this study is to analyze mechanisms that integrate demand
forecasting with replenishment optimization.
Main part. Classical approaches to inventory planning
Traditional inventory planning methods were developed under relatively stable market conditions and assumed
the use of fixed parameters for demand, lead time, and costs. They allowed companies to structure logistics
processes based on deterministic calculations, minimizing expenses and maintaining the required level of
customer service. Although many of these approaches remain relevant today, their effectiveness under conditions
of high uncertainty is gradually decreasing.
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025
Page 1308
Among the most common tools, three main directions can be identified. Each of them is aimed at achieving
specific management goals, but it has both strengths and limitations (table I).\
Table I Classical Methods of Inventory Planning 1, 2
Method
Main purpose
Advantages
Limitations
Economic
order size
Determining the optimal
order volume.
Ease of use, cost
minimization.
It is based on the assumption of
stable demand and unchanged
conditions.
ABC/XYZ
analysis
Classification of the
assortment by importance
and predictability.
Allows you to highlight
priority positions and
distribute attention.
Results depend on criteria,
limited flexibility.
Safety stock
Compensation of
uncertainty of demand and
supply.
Reducing the risk of
shortage, maintaining the
level of service.
The possibility of overestimating
or underestimating the stock
level, increasing costs.
Thus, in conditions of sharp fluctuations in demand and instability of supply, models that were initially optimal
become a source of systematic errors. This limitation served as a prerequisite for the development of intelligent
planning methods capable of taking into account the dynamics of external factors and promptly correcting
decisions.
Predictive analytics and machine learning in demand forecasting
One of the areas of improvement of inventory management is the introduction of predictive analytics. Compared
to retrospective approaches, it is directed towards uncovering hidden patterns in consumption dynamics and
allows more accurate predictions. It is the foundation of adaptive planning systems as it enables consideration
of not only historical data but also a wide range of exogenous variables influencing demand.
The traditional foundation is represented by time series methods. They are based on the assumption that future
demand behavior depends on its past values and that the data contain patterns that can be mathematically
modeled. The standard autoregressive integrated moving average (ARIMA) model allows for the factoring in of
autocorrelation and trend detection, with the seasonal variant SARIMA being used when there is clear cyclicality
present. Other exponential smoothing methods are used to seasonally adjust forecasts based on trend and season
components. Their strengths are stability and mathematical rigor, and limitations manifest themselves with
sudden changes in consumer behavior or under the influence of exogenous factors that are not included in the
model.
The next stage of development was represented by regression models, which make it possible to complement
time series with external variables. This approach is particularly valuable when demand is influenced not only
by internal patterns but also by macro- and micro-environmental factors 3. Single and multiple linear regression
allow for these variables to be controlled and the contribution of each to the final prediction decided. One
advantage of this approach is interpretability, as managers have the ability to observe the direct effect of the
individual variables and make more tactical judgments. With many factors involved, risks of overfitting,
multicollinearity, and reduced model stability exist.
The growth in computing power allowed the change to machine learning algorithms and neural networks,
which provide greater accuracy and the ability to learn nonlinear relationships. In demand forecasting, recurrent
neural networks (RNN) and their advanced modification, long short-term memory (LSTM), are particularly
widespread, as they effectively process long sequences of data. These models can identify complex consumption
patterns and account for time-delayed effects. Additionally, ensemble methods such as Random Forest and
gradient boosting are used, which automatically determine factor importance and minimize errors on large
datasets. The limitations of neural network models include high computational complexity, the need for large
volumes of high-quality data, and low interpretability of results.
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025
Page 1309
A comparative analysis of these approaches shows that each of them has its own area of optimal application.
However, the most promising direction of development is a hybrid approach, in which classical methods are
used as the basic level of forecasting, while intelligent methods increase accuracy and adaptability. This
combination makes it possible to combine the interpretability and transparency of traditional models with the
high predictive power of machine learning methods.
Adaptive inventory planning systems
Demand forecasting methods form the foundation for the transition from static management to more flexible and
intelligent systems. However, even the most accurate forecasts lose their practical value if they are not integrated
into mechanisms of operational inventory management. Therefore, an important stage in the evolution of
management has been the development of adaptive planning systems capable of predicting future needs. These
systems can dynamically adjust procurement and replenishment parameters in response to changing conditions.
The concept of data-driven adaptive management implies abandoning rigidly fixed rules in favor of decisions
generated by algorithms in real time 4. They use streaming analytics that update model parameters in real time
based on incoming data on demand, supply, prices, and customer behavior. This allows for prompt response to
deviations from forecast values and minimizes risks of shortage or overstocking. Unlike traditional approaches,
these systems are «living» models that evolve along with the external environment (Fig. 1).
Fig. 1. Architecture of an adaptive inventory planning system
Their effectiveness is achieved through the integration of demand forecasting with procurement optimization
models. This represents one of the central elements of modern logistics strategy, as it is at this stage that the
balance between storage costs, shortage costs, and logistics expenses is established. The formulation of the
optimization problem in inventory management is expressed as the minimization of total costs while maintaining
the required level of service.
The issue is compounded by the supply uncertainty and the demand being probabilistic in nature. Similarly,
delivery time variability, supply disruption, and price fluctuation have a huge influence on the final decision.
Thus, the traditional models are no longer sufficient, and interest is turned to intelligent algorithms (Fig. 2).
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025
Page 1310
Fig. 2. Methods of intelligent inventory replenishment optimization
One of the most widely used tools in inventory management is stochastic modeling, which makes it possible to
account for the probabilistic distribution of demand and supply parameters. Differently from deterministic
models, this approach allows for generating a sequence of scenarios reproducing real market volatility. For
example, to forecast acquisitions of a pharmacy chain or grocery supermarket, it is possible to factor in the
probability of seasonal peaks in demand, supply lags, or price instability of raw materials . Using this approach,
it becomes feasible to spot shortage and overstock risks at an early stage and select a replenishment strategy best
resilient to unexpected events.
Equally important are evolutionary algorithms used to find optimal solutions in complex multidimensional
inventory management problems. In practice, this is evident in the need to simultaneously determine order
volumes for hundreds of product items with different turnover rates, costs, and shelf lives. Traditional analytical
methods often prove inefficient under such complexity. In contrast, genetic algorithms and the particle swarm
optimization method can find near-optimal solutions even with incomplete information. These algorithms
«learn» from numerous possible order combinations, gradually identifying the most rational options in terms of
minimizing total costs.
In recent years, special attention has been given to the application of reinforcement learning (RL). It views
inventory management as a dynamic process consisting of a sequence of decisions made over time 5. The
algorithm operates in an environment where the state is defined by the current inventory level, forecasted
demand, and supply constraints. Each decision regarding order quantity or timing is accompanied by an
evaluation of its consequences. The system is «rewarded» for taking those actions that minimize total costs and
prevent shortages, and «penalized» for making those decisions which lead to overstocking or lost sales.
This makes it possible to develop a management plan step by step that includes consideration of not only short-
term benefits but also long-term outcomes. For example, under high seasonality conditions, an RL algorithm can
identify that increasing inventory levels ahead of peak demand is beneficial. This adjustment produces a stronger
positive effect than short-term savings on storage costs.
The implementation of RL models imposes increased requirements on data quality and computational resources.
Effective training is possible only with large volumes of historical data on demand, supply, and costs, structured
by time intervals and cleaned of noise. The completeness and representativeness of the dataset determine the
algorithm’s ability to correctly evaluate the consequences of actions and develop a stable strategy. The
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025
Page 1311
computational load is also significant. Training RL requires repeated simulation of environmental states and
optimization of policy parameters. This makes it necessary to use parallel computing, graphics processors, and
cloud infrastructures necessary.
Another essential tool of inventory management is simulation modeling, which offers the potential for
replicating the functioning of a supply chain in virtual space and analyzing the impact of various managerial
decisions. Simulation takes into account myriad variables – from variation in demand during different seasons
and supply delays to warehouse holding capacity constraints and limitations of the transportation infrastructure.
This renders the method particularly valuable in policy-making when designing replenishment policies because
one is able to examine potential outcomes before applying the changes to the real system.
A comparison of the considered methods shows that each of them has its own area of optimal application and
level of maturity. For inventory management systems operating under high uncertainty, the greatest effect is
achieved through their combined use (table II).
Table II Comparative Analysis of Intelligent Inventory Management Methods
Method
Advantages
Typical applications
Stochastic
Accounts for demand and
supply variability; supports risk
assessment.
Seasonal inventory
planning; supply risk
analysis.
Evolutionary
Handles nonlinear,
multidimensional problems;
finds near-optimal solutions.
Procurement
optimization; complex
logistics tasks.
RL
Learns from experience; adapts
to changing environments.
Real-time inventory
control; adaptive
pricing.
Simulation
Enables safe scenario testing;
models system interrelations.
Strategic logistics
planning; network
design.
Thus, the use of these methods enables the integration of forecasting, optimization, and experimental modeling
into a unified decision support system. The practical effectiveness of such approaches is confirmed by the results
of their implementation in corporate logistics systems of large companies. For instance, Walmart clearly uses
stochastic modeling to represent probabilistic distributions of demand as well as delivery delays. Through that,
one is able to construct procurement scenarios that are resilient to region and seasonally caused demand
fluctuations and peaks in consumption. In addition, the company uses simulation modeling of distribution points
and transportation routes so that it is capable of simulating the effects of new supplier addition or changes in
delivery schedules in advance. Their application has contributed to reduced transportation costs and improved
replenishment forecast accuracy, thereby improving the overall supply chain resilience [6].
Another example is that of Amazon, which has integrated RL methods into its warehousing inventory and
distribution center operations systems. Through this, the company has managed to increase the accuracy of
demand forecasted and optimize supply routing between warehouses [7].
The experience of implementing intelligent algorithms in large corporations makes it possible not only to
confirm the practical viability and maturity of these technologies but also to evaluate them based on objective
quantitative indicators. In the context of inventory management, several of the most informative metrics can be
distinguished. One of them is the service level, which reflects the proportion of demand satisfied without
shortages. It serves as an indicator of the system’s ability to respond promptly to market changes and maintain
stable customer supply.
Another important indicator is the order fill rate, which characterizes the responsiveness of order fulfillment and
the efficiency of logistics processes. It is closely related to the quality of demand forecasting and the optimization
of supply routes.
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025
Page 1312
The next metric is the total cost of inventory ownership, which includes procurement, storage, transportation,
and administrative expenses. A decrease in this indicator reflects an increase in the economic efficiency of the
system and a more rational allocation of resources.
Finally, supply chain reliability reflects the resilience of the logistics system to external disruptions. A high
value of this metric indicates the company’s ability to maintain stability and service levels even under unstable
external conditions.
An analysis of these indicators shows that the use of intelligent methods leads to an increase in service level, a
reduction in total costs, and fewer failures and delays. However, despite the clear advantages, their
implementation in real inventory management processes remains a complex task. The main limitations are
associated with the strong dependence on data quality and completeness, the need for significant computational
resources, and the difficulty of integrating algorithms into existing corporate information systems. In addition,
the interpretability of decisions generated by neural network models remains limited, which complicates their
use in industries that require transparent justification of managerial actions.
Future research should focus on developing unified data architectures, adaptive algorithms with reduced
computational demands, and explainable artificial intelligence tools capable of ensuring trust and verifiability of
analytical results. In the long term, it is precisely the combination of the transparency of classical approaches
and the learning capability of intelligent systems that will define the development of sustainable, scalable, and
ethically responsible solutions in inventory and logistics management.
CONCLUSION
Modern supply chains have conditions of uncertain demand, price volatility, and logistics risks, and such
conditions make classical methods of inventory management less and less efficient. Classical models, based on
fixed parameters and past records, are not able to give a proper response to a dynamic outside world, leading to
the increasing cost of operation and a decrease in the level of service. Intelligent techniques permit these
limitations to be relaxed.
Predictive analytics and machine learning algorithms provide more accurate forecasts of consumption dynamics,
while adaptive planning systems based on these methods generate procurement and replenishment decisions that
reflect real environmental changes. The integration of forecasting with optimization models creates the
foundation for flexible and efficient inventory management.
Hence, intelligent data analysis methods are designing a new adaptive planning model that makes it possible for
companies to reduce risk and enhance supply chain efficiency simultaneously. Their use gives scope for pre-
emptive management and reactive scope for subsequent research in terms of developing hybrid solutions that
leverage the interpretability of legacy models combined with the predictive strengths of artificial intelligence.
REFERENCES
1. Drogunova, Y. (2025). The impact of software quality assurance practices on the competitiveness of the
technology sector. International Journal of Advanced Research in Science, Communication and
Technology, 5(1), 735–740.
2. Verma, P. (2024). Transforming supply chains through AI: Demand forecasting, inventory management,
and dynamic optimization. Integrated Journal of Science and Technology, 1(3), 1–14.
3. Roilian, M. (2025). Architectural approaches to building scalable IT systems for supply chain
management. Cold Science, 17, 28–38.
4. Nookala, G. (2024). Adaptive data governance frameworks for data-driven digital transformations.
Journal of Computational Innovation, 4(1), 1–20.
INTERNATIONAL JOURNAL OF RESEARCH AND SCIENTIFIC INNOVATION (IJRSI)
ISSN No. 2321-2705 | DOI: 10.51244/IJRSI |Volume XII Issue X October 2025
Page 1313
5. Rolf, B., Jackson, I., Müller, M., Lang, S., Reggelin, T., & Ivanov, D. (2023). A review on reinforcement
learning algorithms and applications in supply chain management. International Journal of Production
Research, 61(20), 7151–7179. https://doi.org/10.1080/00207543.2022.2140221
6. Roy, H. N., Almehdawe, E., & Kabir, G. (2024). A two-stage stochastic optimization framework for
retail supply chain modeling with contemporaneous resilient strategies. Production Engineering, 18(6),
903–924. https://doi.org/10.1007/s11740-024-01279-x
7. Yang, Y., Wang, M., Wang, J., Li, P., & Zhou, M. (2025). Multi-agent deep reinforcement learning for
integrated demand forecasting and inventory optimization in sensor-enabled retail supply chains.
Sensors, 25(8), 2428.