0% found this document useful (0 votes)
35 views9 pages

Predictive Insights Real-Time Decision Making in Supply Chain Management

logistics

Uploaded by

zhxu7578
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
35 views9 pages

Predictive Insights Real-Time Decision Making in Supply Chain Management

logistics

Uploaded by

zhxu7578
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 9

Proceedings of the International Conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud) (I-SMAC 2024)

IEEE Xplore Part Number: CFP24OSV-ART; ISBN: 979-8-3503-7642-5

Predictive Insights: Real-Time Decision Making


2024 8th International Conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud) (I-SMAC) | 979-8-3503-7642-5/24/$31.00 ©2024 IEEE | DOI: 10.1109/I-SMAC61858.2024.10714831

in Supply Chain Management


Dr. Murugan Ramu Dr. S. Selvaraju
Associate Professor Dr. S. Thangamayan Associate Professor
Department of Management Studies Associate professor Department of Commerce,
Saveetha School of Law Saveetha School of Law, Saveetha School of Law,
Saveetha Institute of Medical and SIMATS, Chennai -77 Saveetha Institute of Medical and
Technical Sciences (SIMATS), Chennai- [email protected] Technical Sciences (SIMATS)
77 Chennai-77
[email protected] [email protected]

Abstract— The supply chain management (SCM) is an important I. INTRODUCTION


aspect to maintain the flow of business activities; it involves good
management and coordination of the network comprising complex
and interconnected parties who are involved in the provision of Supply chain management attempts to get the proper products
products and services to the final users the production, in the proper quantities to the proper location at the suitable time.
transportation, and delivery of the ready goods. The global Through the utilization of effective supply chain management, the
integration of supply chains and the quickening stage of firm not only is able to manage their suppliers, customers and
complexity and dynamic-environment requires the use of warehouses as necessary but is also able to ensure that the
advanced tools and techniques to improve the processes of organization maintains a competitive edge in the supply chain
decision-making and to find the best operational routes. This has world [1] [2]. That the sector is serious about cutting out
made the implementation of advanced machine learning inefficiencies is evidenced by this. At the time, with access to
algorithms relatively more important in supply chain management more technology and a better understanding of all shared roles
(SCM) for improving decision-making processes and optimizing within the supply chain, we were beginning to optimize decisions
operations. In this work, we propose a new model, called Hybrid regarding both demand and supply and their interdependencies.
LeNet-Xception Model (HLXM), which is a hybrid model of Digital technology is changing consumer behavior, and the
LeNet-5 and Xception to achieve a quick and smooth SCM in real- functioning businesses and economic sectors are beginning to
time. These performance comparisons were conducted with both depend on one another [3].
traditional, and state-of-the-art machine learning models, resulting
in improved demand forecasting, inventory handling, logistics The development of the supply chains (SCs) has increased the
optimization, and decision-making productivity in general. size and the complexity of the SCs while globalization and
Results demonstrate the ability of HLXM to materially improve outsourcing has made SCs more exposed to disruptions. Supply
business operations, lower costs, and improve customer chains need to be innovative to keep pace with the more and more
satisfaction turbulent but highly heterogeneous world of the market [4] [5]. As
customization grows, inventory decreases and international
Index Terms— Hybrid LeNet-Xception Model, HLXM, Supply competition intensifies, economies of scale in production become
Chain Management, Inventory Management, Logistics harder to obtain, meaning it is more difficult to control costs and
Optimization, Operational Efficiency, Machine Learning, Deep quality in the supply chain. A number of disruptions can trigger a
Learning, Lenet-5, Xception. circular chain of disruptions that lead to critical SC failures.
Anything from power outages to system crashes, network outages,
and unexpected events [6].

Technology is critical in enabling agility in decision-making.


For instance -This can be seen in impact of AI systems on agile

979-8-3503-7642-5/24/$31.00 ©2024 IEEE 1169


Authorized licensed use limited to: Consortium - Saudi Arabia SDL. Downloaded on October 31,2024 at 16:18:46 UTC from IEEE Xplore. Restrictions apply.
Proceedings of the International Conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud) (I-SMAC 2024)
IEEE Xplore Part Number: CFP24OSV-ART; ISBN: 979-8-3503-7642-5

decision making. To do this, decision-makers use different decision-making in SCM. This paper will seek to illustrate the
evaluations that can be complemented by a proper understanding usefulness of the HLXM for improving different supply chain
of decision-making procedures and their effects [7]. In other management tasks like demand forecasting, inventory
words, and quite simply, clear and easily consumable information management, logistics optimization and supplier performance. A
can accelerate swift decision-making. The high transparency of detailed comparison with traditional and other modern ML models
their operation renders AI approaches particularly useful in demonstrates how the HLXM makes a major difference in making
decision support systems, causing decision takers to be able to act the whole process more operational efficient, less costly and be
rapidly in response to changes [8]. Figure 1 shows the benefits of more customer satisfying.
predictive analytics.
II. RELATED STUDY
Reduced Costs
The best appropriate approach in the Setting of Chilean marine
Faster Data Processing
ports in the Supply Chain Management (SCM) to introduce with
in Supply Chain using the Blockchain Technology (BT) [11]. It
discovers and classifies essential input variables that have an
Risk Prediction
Benefits
impact on the adoption of BT using the Fuzzy Logarithmic
of Method of Additive Weights (LMAW) and Double Normalization
Reduced Downtime
Predictive based Multiple Aggregation Methods (DNMA). The findings
Analytics underscore the importance of effective risk mitigation, more
Boost Production Capacity
robust supply chain processes, and safer transactions. It uses an
MCDA hybrid approach to support decision making under
Extend Equipment Life uncertainty and ports as good examples to integrate BT and the
outcomes of this method contribute to the promotion of more
Improved Resource Management efficient operations to be integrated in all Chilean ports using the
BT for SCM.
Figure 1. Benefits of Predictive Analytics Industry 4.0 leads to huge technical leaps for all links in the
value chain. Previous research has primarily focused on using
Efficiency, speed, quality and flexibility improvements are Industry 4.0 in manufacturing but fallen short in how it impacts on
always in-demand in the industry. And it seems like more and decision-making and adaptation practices to take full advantages
more money is being poured into R&D to develop the smart TC and remedies which is providing data from the apparel and textile
factory of tomorrow, which in turn is having a trickle-down effect industries [12]. Through a systematic literature survey of 61
on the rest of the textile sector [9]. This is all intellectual but research papers published between 2014 and 2019 in Scopus, we
routine work, which in the smart factory of the future will be identified five key decision-making trends: data-driven, real-time,
eliminated to the robotization of work processes. As a result, decentralized, integrated, and sustainable approaches. Among its
working conditions will improve, new jobs will be created and recommendations for the future of the textile industry is
overall efficiency will increase. Unfortunately, the types of addressing problems resulting from digitization, utilization of
demand planning that result from traditional supply chain automation in the context of Industry 4.0 tools, and the need for
management seldom survive the logistical challenges, and demand social as well as environmental responsibility.
vagaries and supply breaks with which global commerce is fraught
[10]. This has caused a wave of interest in addressing such issues The digital era studies of retail supply chains by comparing
utilizing Cutting-edge ML and DL techniques. what managers were responsible for doing before and during
COVID-19, illuminating about how human decision-making has
Hybrid LeNet-Xception Model was considered as a evolved from pre-digital to the digital era in retail supply chains
breakthrough in implementing ML and DL in SCM. HLXM, one [13]. The utility of grounded theory was illustrated through
of these pre-configured models, provides a depth of expertise in application to the 25 in-depth interviews with retail supply chain
supply chain data -with far greater accuracy than the Detection- executives and four key lessons learned are established. Without
ML model - that can capture both simple and complex patterns in doubt, management discretion is an upward trend throughout the
supply chain data. This new hybrid solution makes of both internal supply chain. The second undoubtedly cuts both ways, and the
data (inventory, sales, delivery) that is mixed with external heart is now analytics and judgement together as the consistent
sources like market stats, supplier data, weather forecasts and decision-making premise. Ultimately, the result is the integrative
economic information. This HLXM makes these wide data framework, the end point of this research effort. They identify
sources integrated with each other to solve the issues of real-time what is required for successful analytics and AI rollouts the

979-8-3503-7642-5/24/$31.00 ©2024 IEEE 1170


Authorized licensed use limited to: Consortium - Saudi Arabia SDL. Downloaded on October 31,2024 at 16:18:46 UTC from IEEE Xplore. Restrictions apply.
Proceedings of the International Conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud) (I-SMAC 2024)
IEEE Xplore Part Number: CFP24OSV-ART; ISBN: 979-8-3503-7642-5

methods paves the way for future work in this field to know how Market trends data such as industry-wide sales trends, consumer
to work human judgement alongside analytics to make online preferences, and competitor activities. Information about supplier
shopping decisions correctly. reliability, lead times, lot sizes, and historical performance is
supplier data. Logistics and transportation are severely impacted
The logistics sector is built on the basis of supply chain by inclement weather, which lead to weather forecast predictions.
management, which manages goods from suppliers to the final Economic indicators include GDP growth rates, inflation rates,
customers, enabling the logistics sector to internationalize its and unemployment rates to give perspectives on the broader
capacity for competition and profitability. Digital strategies economy context that impact demand/supply dynamics in supply
enhance decision-making by using the latest technology and chain. Figure 2 shows the architecture of proposed model.
understanding stakeholders more. Civilization around the world
has been flowing through businesses, customers, and economic Data Cleaning
connections have stirred well Gemini technology has flourished.
Artificial Intelligence (AI) renders decision algorithms and Handling Missing Values
process automation, and the mobile Internet makes
communication uninterrupted [14]. Big data is ideal for reviewing Handling Missing Value is as important in data cleaning to
enormous sets, cloud computing lets the exchanging data between make sure about the accuracy and trust-ability of the dataset.
them, and augmented reality supports appreciate of processes. Missing values can handle a number of sources - data entry errors,
Blockchain ensures transactions security, and the Internet of untracked transactions, system errors, etc. For dealing with
Things helps to perform remote system administration. missing values, either they are imputed by their mean, mode
values or they are imputed by prediction. Mean/Median
The Agricultural Supply Chain (ASC) is of great significance imputation - This method involves replacing the missing
in emerging economies like China, Pakistan, and India. When it numerical data with the mean value (mean) or most frequent value
comes to implementing state-of-the-art supply chain optimization (mode) of the feature it belongs to, ensuring that the imputed value
technology, established economies have been far ahead of to represent the central tendency of the data. Mode imputation is
emerging economies like Blockchain and AI. This delay in all used in the case of categorical data. Predictive imputation makes
parts of the supply chain has been the root cause of the non- an educated guess with the help of machine learning models from
transparency, non-efficiency, post-harvest operation problems and the relationships we see among other features. We customize the
non-market participation etc. To identify significant barriers, such features with the data which is even better most of the times as it
as a lack of post-harvest facilities, and the high need for better takes care of data inherent data patterns. With enough thought
market data systems for farmers [15]. They suggested an about the column and how we handled the missing values we
intelligent decision-making framework that overcomes these allow it to be created to a level that is best for the data, ensuring
challenges by introducing traceability in the supply chain and that bias has been minimized and that the dataset remains robust
empowering the farmers in managing the post-harvesting for further analysis.
processes. They found this arises out to provide key information
for not just the entire farming community or the politicians but for
that this can further generate ground for more studies in this
imperative field.

III. METHODOLOGY
Data Source

This data include company-internal information like inventory


balances, sales transactions and delivery times. The stock of
different products at warehouses is shown by inventory levels
hence gives a clear idea of the goods in stock. Sales records
provide a record of the items that have sold in the past along with
quantities sold, as well as when they were sold and to whom.
Delivery schedules contain specifics related to shipment timings,
routes, carrier information and delivery statuses, anything that is
essential for logistics optimization and efficient time-bound
deliveries. The dataset is integrated with third-party datasets from
market trends, suppliers, weather, and macroeconomic data.

979-8-3503-7642-5/24/$31.00 ©2024 IEEE 1171


Authorized licensed use limited to: Consortium - Saudi Arabia SDL. Downloaded on October 31,2024 at 16:18:46 UTC from IEEE Xplore. Restrictions apply.
Proceedings of the International Conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud) (I-SMAC 2024)
IEEE Xplore Part Number: CFP24OSV-ART; ISBN: 979-8-3503-7642-5

Data Source
Normalizing and Standardizing Data

Data Normalizing and Standardizing the data is one of the most


Cleaning
important steps we should perform in order to bring all the features
to a common scale, which will help the model to train more
Handling Missing Values effectively and fit to standard results. Normally normalization
scales the data to a fixed range of [0, 1] where it becomes
Removing Duplicates and Outliers incredibly useful when the data distribution is not Gaussian and
training a network to perform better with normalized input.
Normalizing and Standardizing Data
Standardization means centering the data to have a mean of zero,
and rescaling the feature values to have a standard deviation of
one, which is useful when our input variables have a Gaussian
distribution as shown in the first step which we covered with linear
Feature Extraction
and PCA algorithms. This process helps in taking care of the
features having different units and scales thereby allowing all the
Creating New Features features to contribute equally towards the model. These scaling of
the data normalizes and/or standardizes the data which in turn
Using Domain Knowledge
more than improves the model performance and decrease the time
it takes the model to converge, hence making more precise and
reliable predictions.
Hybrid Model Architecture: Hybrid LeNet-Xception Model (HLXM)
LeNet
Convolution
Feature Engineering
Convolution Convolution
Subsampling (5x5)
(5x5) Subsampling (5x5)

Creating New Features

Feature Map
Feature Engineering does not only mean creating new features;
Input
Feature Map
Feature Map Feature Map
this is a very important value that adds a lot of old insights, and
new patterns from the model. Moving averages have been around
Xception Model
for a long time and are used to smoothen out the short-term clutter
and point out the long-term trends or cycles in time-series data. A
moving average of sales data helps to smooth out the data and can
Hybrid Stratergy
reveal an underlying trend - the noise is reduced. Lagged variables
refer to using previous values of a feature to predict the current or
future values. Good use of lagged variables - previous sales
Figure 2. Architecture of Proposed Model figures are often therein used to predict future sales programmed
for seasonality and commonality in box management. Interaction
Removing Duplicates and Outliers terms are the variables that are constructed by combining two or
more variables together to capture the combined effect of those
Duplicates are basically the same data point being recorded variables on the dependent variable.
multiple times, which could wrongly project in the analysis and
may switch conclusions to another direction. Duplicate detection Using Domain Knowledge
and deduplication is all about comparing data points against the
rest and removing redundant elements so that one date point is the Converting raw data into more advanced features based on
only one for one record. At the same time, outliers are values that domain knowledge is a crucial part of feature engineering which
are tremendously out of the rest values. They can occur due to in turn play a key role in shaping the data in a way that is useful
errors in inputting data, errors in measurement, or simply and informative for the model. However, domain experts can give
infrequent events. Outliers in database may cause statistical us valuable hints about what are the most important features that
analysis to skew for models to get trained on wrong data, due to affect the outcomes and are there any interactions between them.
which wrong predictions will follow. For example, in supply chain management, domain knowledge
can be employed in identification of relevant variables like lead
times, order quantities, supplier reliability, and seasonality effects.
Understanding the business context helps to focus on some of the
more complex parts of the supply chain process and this is what

979-8-3503-7642-5/24/$31.00 ©2024 IEEE 1172


Authorized licensed use limited to: Consortium - Saudi Arabia SDL. Downloaded on October 31,2024 at 16:18:46 UTC from IEEE Xplore. Restrictions apply.
Proceedings of the International Conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud) (I-SMAC 2024)
IEEE Xplore Part Number: CFP24OSV-ART; ISBN: 979-8-3503-7642-5

drives the creation of appropriate features. Similarly, customer


demand elasticity can also be driven to understand how demand is The HLXM substantially improves on-time decision making in
driven with respect to the changes in price or promotions. SCM by utilizing modern deep learning concepts for solving a
variety of problems and for toward optimizing several segments
Hybrid Model Architecture: Hybrid LeNet-Xception Model of the supply chain. The versatility of the HLXM allows the model
(HLXM) to merge the simplicity of the LeNet-5 model and the deepness of
the Xception model, capturing both simpler as well as more
The Lenet-5 is one of the earlier convolutional neural network complex patterns within the dataset, always leading to the correct
architectures that is now only typically used for small dataset and predictions of future demand, inventory levels, and potential
simple tasks due to its relatively simplicity and efficiency. This disruption.
architecture starts with an input layer that reshapes the data as
needed to match the input dimensions of the model. This is a step The HLXM also optimizes the logistics and delivery process by
in the preprocessing process that makes the input data can enter assessing delivery schedules, shipment timings and real-time
the structure of the network. Convolutional Layers, after the input tracking data for lower delivery costs and unmatched delivery
layer is processed through filters such as edges and textures. These times. The model also supports in better supplier management and
layers are designed to find spatial hierarchies in the data. Pooling risk assessment, looking at the suppliers' data and help to identify
layers are used to scale down the size and dimensionality of the some risks as well as propose some alternative suppliers or
feature map to decrease the computational complexity. These mitigation plans. Insights generated by the HLXM can feed into
layers are associated with down sampling operation which is the decision support systems which will deliver interactive, intuitive
process of reducing the size of the feature maps and preserving recommendations and visualizations enabling rapid and informed
essential information while removing less relevant information’s. decision-making. Routine processes can be managed by
automated decision-making modules thereby increasing the
Xception is an extension of the Inception architecture, which is effectiveness. The approach involves ongoing model monitoring
designed to capture more complex patterns in large-scale, diverse and training with new data, making sure it stays up to speed and
image datasets. At the topmost level is the input flow, a collection accurate.
of initial layers which down-sample the input data while retaining
important patterns. This process improves the lower IV. RESULTS AND DISCUSSIONS
computational load and allows more feature extraction. The
middle layers contain many more convolutional layers, which Integrating internal company data such as inventory, sales, and
perform feature extraction over hidden patterns inside the image. delivery data with external sources including market trends,
These layers use depth wise separable convolutions, which are a supplier data, weather forecasts and economic indicators offers a
more efficient version of conventional convolutions that comprehensive perspective for managing inventory and
performable to learn more complex patterns and relationships optimizing logistics. They are used to record sales, which help in
directly in the data. The exit flow containing the last layers of the estimating the sales such as product movement and customer
network, which take in the previously extracted features to ensure behavior. Delivery schedules guarantee necessary on-time
classification or regression- and perform the final stages of the logistics, essential for smooth functioning. Market Context and
process. Layers in this part are used for the model to learn how to Supplier Performance External data provides insights on a market
identify important features for prediction. context as well as supplier performance metrics necessary to make
sound strategic decisions. Various economic statistics like GDP
Hybrid Strategy growth rate, Inflation rate, Unemployment rate provide clear
vision of economic trends and effect supply chain mechanisms.
HLXM concatenates both LeNet-5 and Xception output to
combine the best of both worlds. Those extracted features are It is important to deal with the missing values to keep the dataset
fused by this strategy, leading to the complete representation of intact and also to ensure the correctness of the dataset. The missing
input data. The concatenated layer is the layer designed to data can be addressed using methods like mean/mode imputation
concatenate the final layers of both LeNet-5 and Xception and or predictive modelling which are good at handling the missing
links them to the final point at which we will expect a single data in the dataset. Mean/mode imputation: This means that they
feature map. The model combines the simplicity and efficiency of will fill missing values using the average or most frequent value
LeNet-5 while capturing complex patterns that Xception (mode) in the column again, ensuring that the dataset maintains its
identified. For these final decision-making tasks, extra fully structure. Predictive imputation, that imputes the missing values
connected layers are stacked on top of the concatenated layer. based on the observed data using machine learning models, which
These layers distribute the joined features to a set of operations to helps to reduce the bias and increases the reliability of the analysis.
form the final output suppose classification or regression.

979-8-3503-7642-5/24/$31.00 ©2024 IEEE 1173


Authorized licensed use limited to: Consortium - Saudi Arabia SDL. Downloaded on October 31,2024 at 16:18:46 UTC from IEEE Xplore. Restrictions apply.
Proceedings of the International Conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud) (I-SMAC 2024)
IEEE Xplore Part Number: CFP24OSV-ART; ISBN: 979-8-3503-7642-5

Data Quality and Consistency are addressed with De- quantities, supplier reliability, seasonality effects etc. Being aware
duplication and outlier removal. Analyzing duplicate & outliers of the business context is what leads to building the features
preventing skewed analysis during model building when the catering to supply chain processes intricacies. For example,
model is trained for output predictions. Normalization & knowing the variability of supplier lead time helps to determine
Standardization: These techniques help the data to often scale its and manage supply chain risks. Customer demand elasticity
feature that in turn, helps in training fast of the model. Feature characteristics reflect price or promotional sensitivity. By doing
engineering can help to extract new features that can represent so, the model is forced to incorporate these specific domain
complex relationships and the creation of features is known to knowledge considerations which naturally encode practical and
enhance the predictive capability and can help increase the real-world knowledge resulting in accurate and actionable
accuracy and predictability of the supply chain models. predictions.

Table 1. Demand Forecasting Accuracy Table 2. Inventory Management Efficiency

Forecasted
Forecasted Stock Stock
Actual Demand Improvemen Stockouts Stockouts
Date Demand Product Level Level
Demand (Traditional t (%) Before After
(HLXM) ID Before After
Model) HLXM HLXM
HLXM HLXM
1/1/2024 500 450 495 10 4/10/1900 1000 800 5 2
1/2/2024 600 580 590 1.72 4/11/1900 500 450 3 1
1/3/2024 700 670 695 3.73 4/12/1900 1200 1000 4 1
4/13/1900 800 700 6 2
1/4/2024 800 750 790 5.33 4/14/1900 900 850 2 1
1/5/2024 650 630 645 2.38
The effect of the HLXM model on the efficiency of inventory
Accuracy comparison of demand forecasting in Table 1 and management is shown in Table 2 and Figure 4 for five different
Figure 3 for traditional with HLXM model for five days in early product IDs. Stock levels pre- and post-HLXM are decreased
2024. The proposed HLXM model always surpasses the showcases lower than median inventory post-implementation In
conventional model, with increase by 1.72% to 10%. Such the second case, stockouts go way down once HLXM is
findings underscore the importance of incorporating state-of-the- implemented, thus enhancing inventory control and minimizing
art machine learning techniques such as HLXM on demand supply disruptions. The results indicate the success of more
forecasting as it can deliver more accurate prediction results advanced ML analytics in bettering inventory management,
compare to traditional approaches. adjusting stock levels, and reducing stockouts, and subsequently,
in improving operational efficiency and customer satisfaction by
offering products that are more available.

Figure 3. Actual vs Forecasted Demand Figure 4. Stock Levels and Stockouts Before and After
HLXM
Domain knowledge can be used to generate useful features from
raw data and enables the model to make sense of the data. Domain LeNet-5 is one of the early convolutional neural network
experts for identifying high impact features and their interactions. architectures capable of handling simple tasks and small data-set
For instance, in the case of supply chain management, domain due to its simplicity and efficiency. The model is starting with the
knowledge would consider factors like lead times, order thing called the input layer and, mainly to reshape the data to fit

979-8-3503-7642-5/24/$31.00 ©2024 IEEE 1174


Authorized licensed use limited to: Consortium - Saudi Arabia SDL. Downloaded on October 31,2024 at 16:18:46 UTC from IEEE Xplore. Restrictions apply.
Proceedings of the International Conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud) (I-SMAC 2024)
IEEE Xplore Part Number: CFP24OSV-ART; ISBN: 979-8-3503-7642-5

the model’s required size. The Convolutional layer uses filters to


carry out features extraction which includes edges and textures.
Pooling layers help reduce dimensionality of input data and
decrease computational requirements for the network while
retaining essential information and discarding unnecessary or
redundant data. Fully connected layers take care of the
classification or regression mapping every neuron in one layer to
every neuron in the next which allows the network to learn non-
linear combinations of features to create very accurate predictions.

Xception is used to extract very complex patterns from huge


datasets. Data processed continues towards the entry flow that Figure 5. Average Delivery Time and Delivery Delays
down-sampled it in a way that kept relevant features of it,
decreasing computational cost. The depth wise separable The HLXM connects LeNet-5 and Xception outputs with a
convolutions are highly efficient, they provide better performance concatenation layer to combine the efficient feature extraction in
and allow to learn intricate features in the deep convolutional the source space layer and the complex pattern recognition in the
layers of the middle flow. In exit flow, features extracted by a target space layer. These features are then processed by fully
backbone are refined for classification or regression, which leads connected layers to increase prediction accuracy and at the same
to accurate predictions. Such architecture of Xception performs time handle computational efficiency.
well in large and complex dataset hence, became a powerful
member of HLXM. Table 4. Supplier Performance Improvement

Table 3. Logistics Optimization On- On- Lead


Lead
Time Time Time Improvem
Time
Delive Delive Variabili ent in On-
Averag Averag Variabili
ID ry ry ty Time
e e Deliver Deliver ty After
Reductio Before After Before Delivery
Deliver Deliver y y HLXM
Deliver n in HLX HLX HLXM (%)
y Time y Time Delays Delays (days)
y ID Delays M (%) M (%) (days)
Before After Before After
(%) S00
HLXM HLXM HLXM HLXM 85 95 3 1 11.76
1
(hours) (hours)
S00
D001 24 20 4 1 75 75 88 5 2 17.33
2
D002 30 25 5 2 60
S00
D003 28 22 3 1 66.67 80 92 4 2 15
3
D004 26 21 4 1 75
S00
D005 32 26 6 2 66.67 70 85 6 3 21.43
4
S00
Table 3 and Figure 5 describes the effect that the HLXM model 78 90 4 2 15.38
5
has with respect to the average delivery times or delays with five
delivery IDs for logistics optimization. The most important metric Table 4 and Figure 6 describes how effective the HLXM is in
reveals an average reduction of delivery time, as per the graph, enhancing the performance of the suppliers, for five random
which declined after the introduction of HLXM, showing delivery supplier IDs. The result after implementing HLXM is an increase
operations became more efficient. Further, the delivery delays by 7-15 percentage points in the proportion of deliveries that are
reduces significantly post-HLXM adoption, to the tune of 60% for on time. Lead time variability also decreases significantly post-
one, to 75% on another. The results illustrate the value of applying HLXM, with improvements seen for 1-4 days. The results show
cutting edge AI in logistics, making speedy deliveries safer. These the model was significantly successful in improving supplier
enhancements play an important role in providing higher service performance for this APC. These improvements are important in
levels and operational efficiency, necessary to satisfy customer ensuring supply chain sustainability, redundancy, or continuity of
demands and allowing for greater supply chain performance. service.

979-8-3503-7642-5/24/$31.00 ©2024 IEEE 1175


Authorized licensed use limited to: Consortium - Saudi Arabia SDL. Downloaded on October 31,2024 at 16:18:46 UTC from IEEE Xplore. Restrictions apply.
Proceedings of the International Conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud) (I-SMAC 2024)
IEEE Xplore Part Number: CFP24OSV-ART; ISBN: 979-8-3503-7642-5

Figure 6. On-Time Delivery and Lead Time Variability Figure 7. Decision Accuracy and Improvement

Table 5. Real-Time Decision Accuracy Table 6. Model Comparison

Accuracy Accuracy HLXM


Metric LSTM XGBoost
Before After Improvement Model
Decision Type
HLXM HLXM (%) Demand Forecasting
88 90 95
(%) (%) Accuracy (%)
Demand Inventory Holding
85 95 11.76 95,000 90,000 80,000
Forecasting Cost ($)
Inventory Stockout Rate (%) 8 7 5
80 92 15
Replenishment Delivery Delay
65 70 75
Supplier Reduction (%)
78 88 12.82
Selection Supplier On-Time
88 90 95
Logistics Delivery (%)
82 93 13.41
Planning Transportation Cost
45,000 42,000 40,000
Overall ($)
81.25 92 13.23
Accuracy Customer
8 8.5 9
Satisfaction Score
Table 5 and Figure 7, the improvement in real-time decision Overall Decision
accuracy with the implementation of HLXM on multiple types of 85 87 92
Accuracy (%)
decisions. Demand forecasting, inventory replenishment, supplier
selection, and logistics planning after HLXM adoption is almost Table 6 shows the performance metrics between LSTM,
11.76% to 15% more accurate. An increase in overall decision XGBoost, and HLXM models for various facets of supply chain
accuracy from 81.25 to 92% suggests that the receiving of this management. HLXM performs the best in almost all the
additional information leads to an enhancement in informed evaluation metrics as expected: it has slightest stock out rate of
decision making across a variety of elements of the supply chain 5%, results in a 95% accuracy in demand predictions, decreases
arena; such findings highlight HLXM's ability to incorporate inventory holding costs to $80,000 and has maximum decrease in
advanced data processing capabilities, improving the quality of delivery delays up to 75%. In addition, HLXM is 95% effective in
IRT predictions and efficiency of operations. Through improving supplier on-time delivery and reduces transportation costs to a
resource allocation, reducing costs, and increasing overall total cost of $40,000.
business flexibility to the ever-changing market and stakeholder V. CONCLUSION AND FUTURE SCOPE
needs, the model will refine decision-making processes. In the supply chain management, the adaption of HLXM yields
a substantial enhancement in the real-time decision making. The
HLXM can increase demand forecasting accuracy, optimize
inventory levels and improve logistics efficiency, as well as
significantly outperform traditional and other modern models due
to the fusion of LeNet-5's powerful feature extraction and
Xception's excellent pattern recognition capabilities. It brings
about many benefits including less inventory holding cost,
minimal stock outs, lower transportation cost and enhanced
customer satisfaction. This comparison, puts the model at a better

979-8-3503-7642-5/24/$31.00 ©2024 IEEE 1176


Authorized licensed use limited to: Consortium - Saudi Arabia SDL. Downloaded on October 31,2024 at 16:18:46 UTC from IEEE Xplore. Restrictions apply.
Proceedings of the International Conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud) (I-SMAC 2024)
IEEE Xplore Part Number: CFP24OSV-ART; ISBN: 979-8-3503-7642-5

demonstration of reduction of supplier fidelity to the buyers and “Mobile edge computing for efficient energy management systems”, book
title, Human-Assisted Intelligent Computing, IOP Publishing Ltd 2023
reduction of delivery delays. In a nutshell, as the HLXM has been
Pages 16-1 to 16-17, 2023 https://doi.org/10.1088/978-0-7503-4801-0ch16.
designed to overcome multiple SCM related hurdles, it provides
an extensive solution as a whole to timely and error free decision [13] Rebekah I. Brau, et al. (2024). "Utilizing people, analytics, and AI for
decision making in the digitalized retail supply chain", JBL, 45, e12355,
making. The process can be further improved to integrate more DOI: 10.1111/jbl.12355
varied and granular input data, such as real-time IOT sensor data
as well as social media sentiment analysis for deeper model [14] R. Mishra, V. V. Desai, R. Krishnamoorthy, M. A. Begum, J. Ranga and S.
N. Taqui, "Energy-Efficient IoT with Deep Learning: Optimizing Resource
performance. Furthermore, future research could investigate Allocation in Smart Grids," 2023 9th International Conference on Smart
whether the performance of SCM systems could be improved by Structures and Systems (ICSSS), CHENNAI, India, 2023, pp. 1-6, doi:
applying reinforcement learning to these systems. Increases in 10.1109/ICSSS58085.2023.10407067.
computational efficiency and scalability will be essential for [15] T. Kalimuthu, et al., (2024), "Intelligent decision-making framework for
implementing the HLXM in larger, more complex supply chains. agriculture supply chain in emerging economies: Research opportunities
and challenges", CEA, Volume 219, 108766, ISSN 0168-1699, DOI:
10.1016/j.compag.2024.108766
REFERENCES
[1] Syed Imran Zaman, et al., (2023), "A grey decision-making trial and
evaluation laboratory model for digital warehouse management in supply
chain networks", DAJOUR, Volume 8, 100293, ISSN 2772-6622, DOI:
10.1016/j.dajour.2023.100293
[2] Surajit Bag, et al., (2023), "Guest editorial: Data-driven quality management
systems for improving supply chain management performance", TQM, Vol.
35 No. 1, pp. 1-4, DOI: 10.1108/TQM-11-2021-315
[3] J. Ranga, T. Arun Srinivas, S. Kumar, H. Anandaram, P. Kulkarni and M.
Amina Begum, "Analysis of Artificial Intelligence based Forecasting
Techniques for Renewable Wind Power Generation," 2022 Smart
Technologies, Communication and Robotics (STCR), Sathyamangalam,
India, 2022, pp. 1-5, doi: 10.1109/STCR55312.2022.10009141.
[4] Sina Nayeri, et al., (2023), "Towards a responsive supply chain based on the
industry 5.0 dimensions: A novel decision-making method", ESWA,
Volume 213, Part C, 119267, ISSN 0957-4174, DIO:
10.1016/j.eswa.2022.119267
[5] Sara Romagnoli, et al., (2024) "The Impact of Digital Technologies and
Sustainable Practices on Circular Supply Chain Management", Logistics.
2023; 7(1):1, DOI: 10.3390/logistics7010001
[6] Jayapal Reddy Vummadi, et al., (2024), "Integration of Emerging
Technologies AI and ML into Strategic Supply Chain Planning Processes to
Enhance Decision-Making and Agility", IJSCM, 9(2), 77–87, DOI:
10.47604/ijscm.2547
[7] Krishnamoorthy, R., Kumar, U., Swathi, G. et al. Metamaterial Inspired Quad-
Port Multi-Antenna System for Millimeter Wave 5G Applications. J
Infrared Milli Terahz Waves (2023). https://doi.org/10.1007/s10762-023-
00921-6
[8] Ghazi M. Magableh, et al., (2024), "A Developed Model and Fuzzy Multi-
Criteria Decision-Making Method to Evaluate Supply Chain Nervousness
Strategies", Mathematics 12, no. 10: 1604, DOI: 10.3390/math1210160
[9] Kiarash Sadeghi R., et al., (2024), "Explainable artificial intelligence and
agile decision-making in supply chain cyber resilience", DSS, Volume 180,
114194, ISSN 0167-9236, DOI: 10.1016/j.dss.2024.114194
[10] Suneel, S., Balaram, A., Amina Begum, M. et al. Quantum mesh neural
network model in precise image diagnosing. Opt Quant Electron 56, 559
(2024). https://doi.org/10.1007/s11082-023-06245-y
[11] Claudia Durán, et al., (2024), "Leveraging Blockchain for Maritime Port
Supply Chain Management through Multicriteria Decision Making",
Mathematics 12, no. 10: 1511, DOI: 10.3390/math12101511
[12] Padmanayaki Selvarajan, Betty Elezebeth Samuel, Krishnamoorthy
Ranganathan, Arvind Kumar Shukla, M Amina Begum and Sundaram Arun,

979-8-3503-7642-5/24/$31.00 ©2024 IEEE 1177


Authorized licensed use limited to: Consortium - Saudi Arabia SDL. Downloaded on October 31,2024 at 16:18:46 UTC from IEEE Xplore. Restrictions apply.

You might also like