See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/375638886

Artificial Neural Network Approach for Inventory Management Problem

Article · November 2023

CITATIONS READS

0 75

1 author:

Randhir Singh Baghel

Poornima University
15 PUBLICATIONS 98 CITATIONS

SEE PROFILE

All content following this page was uploaded by Randhir Singh Baghel on 15 November 2023.

The user has requested enhancement of the downloaded file.

 World Academy of Science, Engineering and Technology
International Journal of Mathematical and Computational Sciences
Vol:17, No:11, 2023

Artificial Neural Network Approach for Inventory

Management Problem
Govind Shay Sharma, Randhir Singh Baghel

 more elaborate mathematical inventory models and by

Abstract—The stock management of raw materials and finished including new algorithms that take into account the dynamics
goods is a significant issue for industries in fulfilling customer and stochastic nature of processes in an inventory system.
demand. Optimization of inventory strategies is crucial to enhancing References [10], [23], and [24] considered a periodic review of
customer service, reducing lead times and costs, and meeting market fully back-ordered order up-to-level (R, S)-system with
demand. This paper suggests finding an approach to predict the
stationary gamma-distributed demand and constant lead time.
Open Science Index, Mathematical and Computational Sciences Vol:17, No:11, 2023 publications.waset.org/10013352/pdf

optimum stock level by utilizing past stocks and forecasting the

required quantities. In this paper, we utilized Artificial Neural Network
The stock-out probability and the fill rate are two service level
(ANN) to determine the optimal value. The objective of this paper is requirements that are handled concurrently and calculated the
to discuss the optimized ANN that can find the best solution for the exact safety factors for different model parameters [10]. The
inventory model. In the context of the paper, we mentioned that the k- behaviour of the resulting order was studied through simulation
means algorithm is employed to create homogeneous groups of items. and appeared to be quite satisfactory. Reference [8] determined
These groups likely exhibit similar characteristics or attributes that the quantity of blood units that must be obtained by hospitals
make them suitable for being managed using uniform inventory and processed by the blood center in order to reduce overall
control policies. The paper proposes a method that uses the neural fit costs as well as levels of shortage and waste in the blood supply
algorithm to control the cost of inventory.
chain. References [1], [25], and [26] presented a careful study
and analysis of the existing manual sales and inventory control
Keywords—Artificial Neural Network, inventory management,
optimization, distributor center.
system of a pharmaceutical store. They aimed to design and
develop a computer-based system to increase the efficiency and
I. INTRODUCTION accuracy of the business operations of the pharmaceutical store.
This work demonstrated the effective and efficient use of a
I NVENTORY optimization techniques can be applied in
industries at the start of the process where raw material
reaches the plant and each level of the process needs to be
computer-based drug sales inventory control system and other
software applications, where an intuitive and flexible user
interface played an important role. References [11] and [14]
maintained. According to [7], [16], and [17], it is only possible discussed the construction of an ANN model that can be used
to minimize the overall cost of the supply chain when every to facilitate the optimization of inventory levels and thus
supply chain link optimizes the base stock level. The problem improve the ordering system and inventory management.
behind this is that the excess stock level and shortage level are Reference [2] investigated the application of ANN in the
not static every time. Reference [20] developed an efficient ordering system of companies. The model was developed to
approach using a genetic algorithm to determine the most optimize the order amount according to the current demand in
probable stock level for inventory optimization in the supply the market. Reference [18] investigated inventory optimization
chain cost. Reference [13] proposed an innovative and efficient in the auto industry using swarm optimization. References [2],
methodology to facilitate the precise determination of inventory [27]-[29] introduced the complete programming of the system
optimization in the supply chain, ensuring minimal total supply in detail using MATLAB 7.0. According to [15], [30]-[32],
chain cost. MATLAB 7.4 was utilized to implement the statistical methods can help identify the best neural network for
proposed approach and evaluate the performance. The genetic reducing inventory levels by up to 50% in any organization.
algorithm performed well, as anticipated. References [13], [18], Reference [5] developed an ANN model to find the optimum
and [19] discussed the application of control theory to the level of finished goods and the levels of holding and carrying
problem of timely acquisition of extra production capacity. inventory.
Reference [12] presented an inventory control algorithm of References [6], [33], and [34] discussed the MFT algorithm
mathematical models of the inventory system. The algorithm to minimize the simulation time compared to other approaches.
was implemented as software modules in the MATLAB Reference [3] analysed research in neural network design and
program [20]-[22]. the optimal values of genetic algorithm operators. Reference [6]
The algorithms provided solutions to basic inventory control presented a review of the state-of-the-art view of the depth and
problems for each type of company. Inventory control breadth of NN optimization through GA searches. References
algorithms and software modules can be developed based on [4] and [35] presented a review of prediction techniques using

Dr. Randhir Singh Baghel is with the Department of Mathematics, Poornima Dr. Govind Shay Sharma is with the Department of Mathematics,
University, Jaipur-303905, Rajasthan, India (e-mail: [email protected]). Vivekananda Global University, Jaipur, Rajasthan, India (e-mail:
[email protected]).

International Scholarly and Scientific Research & Innovation 17(11) 2023 160 ISNI:0000000091950263
 World Academy of Science, Engineering and Technology
International Journal of Mathematical and Computational Sciences
Vol:17, No:11, 2023

ANNs. The technique was used to improve the prediction centroid based on the mean of the assigned data points. This is
accuracy of the model with experimental data. Reference [9] done simply by clustering the zero and non-zero values. They
provided an overview of ANNs, explaining the working and arrange the data (stocks) in a significant manner as a K-mean
training of ANN, as well as its applications and advantages. cluster.

II. PRELIMINARY
In neural fitting, we can compute the approximate solution to
find the best among them. It belongs to a particular class of
evolutionary algorithms that are inspired by biology. It is a
simulation where solutions are represented to determine the
best fit. In this approach, solutions can be encoded using
numeric coding other than binary coding. This paper proposes
a method that utilizes the neural fit algorithm to control the cost
of inventory. Here, we will employ the Neural Cluster Fig. 1 K-mean cluster data
Open Science Index, Mathematical and Computational Sciences Vol:17, No:11, 2023 publications.waset.org/10013352/pdf

Algorithm to maintain the stock level. We are considering k-

clustering and the supply chain length μ, which refers to the In Fig. 1, the most recent data and previous stock levels may
number of members in the supply chain, such as factories, be utilized. The fill rate and the stock-out chance were two
distribution points, and suppliers. We take the three stages of service-level requirements that were handled concurrently. All
inventory factory to distribution centers 1 & 2. of these points are connected to Fig. 1. The inventory sheet is
The factory serves as the manufacturing unit, where a created randomly from the stock levels within the lower and
significant amount of goods is stocked. Inventory optimization upper limits at factory and distribution centers 1 & 2. The initial
techniques can be applied in industries at the beginning of the data of inventory prepared with the raw data taken from the
process when raw materials reach the plant. It is necessary to initial factory, distributor center 1, distributor center 2, and the
maintain each level of the process to ensure smooth operations. supplier point are fixed requirements for further process.
The massive stocks are held here in this area and transported as
per the requirements of other distribution centers. The material III. RESULT AND DISCUSSION
has to first move to distribution center 1 and then to transport to Table I shows the stock levels at distribution centers 1 & 2.
the other center 2. In the supply chain system, we focused on The factory data show the manufacturing at the factory point;
the massive stock holding area and the manufacturing area of distribution point 1 & 2 shows the quantities to be dispatched
material as well as the distribution point. The distribution points to the supplier points or warehouses. The data of the quantities
are placed in the middle of the procedure where we supply the at the supplier end have to be fixed and they have to be
material for shipment. In our study, we take the two distribution maintained for the customers. According to table data the
points and then ship them to the supplier. This chain is very supply point is used to deliver the material as per the demand
critical for inventory control and material handling. Our of the customer. There are experimentally 94 data collected but
method's job is to forecast an ideal stock level using historical we have taken 28 types of observation due to repetition of stock
data, ensuring that there is seldom a shortage and that there is requirement. The output y is determined by the inventory levels
never an excess of inventory using the forecasted stock level. It x1, x2, and x3, and the goal is to optimize them using a neural
follows that our method ultimately determines the minimum network.
stock levels required at the plant, distribution center 1, and A. Experimental Result and Analysis
distribution center 2, the three supply chain participants.
In the beginning, there is an excess or shortage of stock at the Network builders need to experiment with numerous neural
distribution points. There are two choices for representing the networks before finding the right one for the problem at hand,
inventory control by zero and non-zero. Non-zeroes refer to the enabling effective reduction of inventory levels. There is no
requirement of inventory to control. The non-zero data states general theory that specifies the type of neural network, number
both the excess number of stocks as well as the shortage of layers, number of nodes (at various layers), or learning
amount. The excess amount is given as a positive value algorithm for a given problem. The data from Table I will be
and the shortage amount is mentioned as a negative value. The used for the neural network fitting curve as the output target is
first process is to prepare the cluster of stock levels that may be required to maintain each time. For training the process of
in excess or shortage. The k-means algorithm is a popular neural networking we use Table I center data as input and the
unsupervised machine learning technique used for clustering target is fixed. Now we use the neural network in MATLAB for
data points into distinct groups. It works by iteratively assigning training the process.
data points to the nearest cluster centroid and updating the

International Scholarly and Scientific Research & Innovation 17(11) 2023 161 ISNI:0000000091950263
 World Academy of Science, Engineering and Technology
International Journal of Mathematical and Computational Sciences
Vol:17, No:11, 2023
Open Science Index, Mathematical and Computational Sciences Vol:17, No:11, 2023 publications.waset.org/10013352/pdf

Fig. 2 Training Process in ANN

Fig. 3 Fitting Function

International Scholarly and Scientific Research & Innovation 17(11) 2023 162 ISNI:0000000091950263
 World Academy of Science, Engineering and Technology
International Journal of Mathematical and Computational Sciences
Vol:17, No:11, 2023
Open Science Index, Mathematical and Computational Sciences Vol:17, No:11, 2023 publications.waset.org/10013352/pdf

Fig. 6 Error Histogram

TABLE I
DATASHEET OF QUANTITIES FOR OPTIMUM INVENTORY
Input Target
S.No Distributor Distributor Supplier point
Factory
center-1 center-2 (app. 200)
1 -591 -329 269 200
2 -479 -796 -548 210
3 -591 -329 269 202
4 494 392 285 206
5 -591 -329 269 209
6 372 573 -345 196
7 999 -934 108 205
Fig. 4 Neural Network Performance 8 146 118 532 211
9 -591 -329 269 231
10 -591 -329 269 185
11 -746 721 -677 186
12 792 -456 837 176
13 -591 -329 269 190
14 -550 -634 158 181
15 611 -295 -443 161
16 497 -170 847 157
17 -992 268 -270 206
18 162 969 -507 203
19 482 -471 761 191
20 -591 -329 269 176
21 671 -768 -367 177
22 446 -916 129 202
23 -278 582 -573 201
24 -758 -766 367 183
25 -208 -306 403 173
26 -437 -850 407 177
27 -591 -329 269 191
28 297 342 686 182

The real data from industries can be used to establish and

validate the neural network model. Training the neural network
processes the experimental data into input-output data sets that
are fed to the network. The neuron gives a signal to the neurons
Fig. 5 Output Validation
in the adjacent layer through a non-linear activation function.
The training epoch set for each neural network is 10,000, and it
was trained using the feed-forward technique. Reducing the

International Scholarly and Scientific Research & Innovation 17(11) 2023 163 ISNI:0000000091950263
 World Academy of Science, Engineering and Technology
International Journal of Mathematical and Computational Sciences
Vol:17, No:11, 2023

Mean Squared Error (MSE) is the training's main goal. plankton dynamics with diffusion in a closed system: a pattern formation,
Applied Mathematics and Computation, 218, 17, 2012, pp 8925-8936
According to the suggested neural network's training results, [17] J Dhar, RS Baghel, Role of dissolved oxygen on the plankton dynamics
error (MSE.0001) is reached after 65 epochs. To identify the in the spatiotemporal domain, Modeling Earth Systems and Environment
optimal network, many training procedures are evaluated. The 2 (1), 2016, pp 1-6
[18] RS Baghel, J Dhar, R Jain, Bifurcation and spatial pattern formation in
network's transfer function is modified and tested at the hidden spreading of disease with incubation period in a phytoplankton dynamics,
and output layers during the design stage. We used the training Electronic Journal of Differential Equations 2012 (21), 2012, pp1-12
Levenberg-Marquaredff algorithm for performance MSE. With [19] RS Baghel, J Dhar, Pattern formation in three species food web model in
a coefficient of determination (R2) and Mean Absolute spatiotemporal domain with Beddington–DeAngelis functional response,
Nonlinear Analysis: Modelling and Control 19 (2), 2014, pp 155-171
Percentage of Error (MAPE) value between the model [20] RS Baghel, J Dhar, R Jain, Chaos and spatial pattern formation in
prediction and actual values of 0.9999 and 0.0001, respectively, phytoplankton dynamics, Elixir Applied Mathematics 45, 2012, pp 8023-
over 6 samples, the neural network performed well. 8026
[21] RS Baghel, J Dhar, R Jain, Analysis of a spatiotemporal phytoplankton
dynamics: Higher order stability and pattern formation, World Academy
IV. CONCLUSION of Science, Engineering, and Technology 60, 2011, pp1406-1412
[22] RS Baghel, J Dhar, R Jain, Higher order stability analysis of a spatial
An ANN model can be developed for industrial inventory phytoplankton dynamics: bifurcation, chaos and pattern formation, Int J
Open Science Index, Mathematical and Computational Sciences Vol:17, No:11, 2023 publications.waset.org/10013352/pdf

control of finished goods as well as raw materials. It can be used Math Model Simul Appl 5, 2012, pp113-127
to optimize the setup cost, material handling cost, and carrying [23] RS Baghel, Dynamical Behaviour Changes in Response to Various
Functional Responses: Temporal and Spatial Plankton System, Iranian
cost of the inventory. The multilayer feed-forward ANN of 4 Journal of Science, 47, 2023, pp1-11
inputs and 10 hidden layers of neurons provides an output [24] J.Dhar, M. Chaudhary, R.S.Baghel and A.C. Pandey, 2015 “Mathematical
which compared with the target. The paper showed the result of Modelling and Estimation of Seasonal Variation of Mosquito Population:
A Real Case Study,” Bol. Soc. Paran. Mat., vol. 33 2 (2015): 165–176.
the ANN model and proved that the model has provided a better [25] O.P. Misra, R. S. Baghel, M. Chaudhary and J.Dhar, 2015
optimistic result for inventory control. “Spatiotemporal based predator-prey harvesting model for fishery with
Beddington-Deangelis type functional response and tax as the control
entity,” Dynamics of Continuous, Discrete and Impulsive Systems Series
REFERENCES
A., vol. 26 2 (2019): 113--135.
[1] Eme O., Uchenna Ugboaja C.A., Uwazuruike F.O., &Ukpai C.U. (2018). [26] S. Pareek, RS Baghel, Modelling and Analysis of Prey-Predator
Computer–based drug sales and inventory control system and its Interaction on Spatio-temporal Dynamics: A Systematic, 4th International
applications in pharmaceutical stores. I.J. Education and Management Conference On Emerging Trends in Multi-Disciplinary Research
Engineering, 30-39. “ETMDR-2023”,77
[2] Guo C., & YangX. (2011). A programming of genetic algorithm in [27] Kaushik P, Baghel RS, Khandelwal S, (2023) The Impact of Seasonality
Matlab7.0. Modern Applied Science, 230-235. on Rainfall Patterns: A Case Study, International Journal of Mathematical
[3] HarunaC., Abdulkareem S., & Hermawan A. (2017). Neural networks and Computational Sciences Vol 17 (10), pp 138-143
optimization through genetic algorithm searches: A review. Applied [28] Baghel RS, Sharma GS, (2023) An Ecological Model for Three Species
Mathematics & Information Sciences, 1543-1564. with Crowley–Martin Functional Response, International Journal of
[4] Mhatre M.S., Dongre M., & Thakur P. (2015). A review paper on artificial Mathematical and Computational Sciences Vol 17 (10), pp 138-143
neural network: A prediction techniques. International Journal of [29] Chaudhuri T. D., (2016) Artificial Neural Network and Time Series
Scientific & Engineering Research, 6. Modeling Based Approach to Forecasting the Exchange Rate in a
[5] Paula S.K., & Azeema A. (2011). An artificial neural network model for Multivariate Framework” Journal of Insurance and Financial
optimization of finished goods inventory. International Journal of Management, Vol. 1 (5), pp 92-123.
Industrial Engineering Computations, 431–438. [30] Khashei, Mehdi, and Mehdi Bijari. (2011) "A novel hybridization of
[6] Peterson C., & James R.A. (1988). Neural networks and NP-complete artificial neural networks and ARIMA models for time series forecasting."
optimization problems; A performance study on the graph bisection Applied Soft Computing 11(2), pp 2664-2675
problem. Complex Systems, 59-89. [31] Taskaya-Temizel, T., & Casey, M. C. (2005). A comparative study of
[7] Radhakrishnan P., Prasad V., & Gopalan M. (2009). Optimizing inventory autoregressive neural network hybrids. Neural Networks, 18(5-6), 781-
using genetic algorithm for efficient supply chain management. Journal of 789
Computer Science, 233-241. [32] Bertsimas, D.; Thiele, A. (2006) A robust optimization approach to
[8] Ramírez A.P., & Labadie N. (2017). Stochastic inventory control and inventory theory. Oper. Res. 54, 150–168.
distribution of blood products. Industrial Engineering and Operations [33] Sarimveis, H.; Patrinos, P.; Tarantilis, C.D.; Kiranoudis, C.T. (2008)
Management, 420-430 Dynamic modeling and control of supply chain systems: A review.
[9] Sonali, & Maind B. (2014). Research paper on basic of artificial neural Comput. Oper. Res. 35, 3530–3561
network. International Journal on Recent and Innovation Trends in [34] Lima, R.M.; Grossmann, I.E.; Jiao, Y. Long-term scheduling of a single-
Computing and Communication, 96 – 100. unit multi-product continuous process to manufacture high performance
[10] Strijbosch L.W.G., &Moors J.J.A. (1999). Simple expressions for safety glass. Comput. Chem. Eng. 2011, 35, 554–574.
factors in inventory control. Tilburg: Econometrics, 1-25. [35] Chu, Y.; You, F.; Wassick, J.M. (2015) Agarwal, A. Simulation-based
[11] Šustrová, T. (2016). A suitable artificial intelligence model for inventory optimization framework for multi-echelon inventory systems under
level optimization. Trendy ekonomiky a managementu trends economics uncertainty. Comput. Chem. Eng. 73, 1–16.
and management, 48–55.
[12] Vladimir M., & Tasho T. (2011). A programme implementation of several
inventory control algorithms. Bulgarian Academy of Sciences, 64-78.
[13] White A.S., & Censlive M. (2016). Inventory control systems model for
strategic capacity acquisition. Journal of Industrial Engineering, Article
ID 1650863.
[14] Yadav A.S., Swami A., & Ahlawat N. (2018). A green supply chain
management of auto industry for inventory model with distribution
centers using particle swarm optimization. Selforganizology, 10-18.
[15] Bansal V.K., (2011) Application of geographic information systems in
construction safety planning, International Journal of Project
Management Vol 29 (1), pp 66-77
[16] J Dhar, RS Baghel, AK Sharma, Role of instant nutrient replenishment on

International Scholarly and Scientific Research & Innovation 17(11) 2023 164 ISNI:0000000091950263