Proceeding of International Conference on Information Science and Technology Innovation (ICoSTEC)

Implementation of KOL marketing as a SME’s marketing strategy

2024-02-15T08:27:55+00:00

SMEs are an ironic business sector which in its function is a pillar of the nation's economy, but in its marketing it still experiences many obstacles. Among them is the difficulty of SMEs in developing their marketing strategy. Furthermore, the KOL Marketing strategy is one of the marketing strategies that have recently become a trend to be implemented in many business sectors. KOL marketing is considered more effective in influencing audiences to make purchasing decisions so that it has an impact on rapid sales. Unfortunately, KOL marketing is identically implemented by big brands because it is assumed that KOL marketing costs are relatively expensive, while SMEs themselves are very limited in terms of financing. In the midst of limited funding and poor marketing implementation, Nana Baby Carrier as one of the RKS Sleman SMEs was able to adapt KOL marketing strategies and implement them according to the SMEs conditions. KOL Marketing concepts and strategies are imitated and modified to remain relevant to SME business conditions. So it actually gives birth to a KOL marketing concept that is right on target and more efficient. This research aims to determine the KOL Marketing Strategy in Improving the Product Branding of the SMEs of RKS study on the SMEs Nana baby carrier. KOL marketing itself is a development of digital marketing. This research uses qualitative descriptive research methods and characteristics assisted by interview data collection techniques with two informants (Founder and Cofounder Nana Baby Carrier) and document analysis. The research results show that there are six ways to create a strategy for using KOL to increase brand awareness.

Detection Of Worker Presence Using The Yolo Model Based On Digital Image

2024-03-20T17:05:53+00:00

The large number of workers requires a long time for administrative officers to check worker attendance manually so that work efficiency cannot be achieved. It takes quite a long time to check worker attendance manually. Attendance reports are sent to office social media, including the names of each worker and so on. This is considered less efficient and sufficient as a basis for providing a solution by detecting worker presence using the YOLO (You Only Look Once) method, so that the process of checking worker attendance can be more efficient. The test results with low pixels and training of 100 epochs, namely 224x192 pixels, obtained an accuracy of 86%, while the best test results were using dimensions of 1088x640 pixels in the worker's photo as test data with original photo dimensions of 1080x1920 pixels in the YOLO model which succeeded in detecting faces with 100% accuracy. So, it can be concluded that the higher the pixel value, the better the accuracy tends to be. However, in this case it also has pixel limitations that are recognized by the model.

Securing Data Files Using RSA-ElGamal and Diffie-Hellman Algorithms

2024-03-20T16:57:55+00:00

Data security involves efforts to protect and ensure three key aspects in the cyber realm: the confidentiality, integrity, and availability of data. It assures cyber users that their privacy is safeguarded, whether on personal computers, mobile devices, or during internet browsing activities. The objective of this study is to authenticate data while applying the SHA-256 function in cryptographic algorithms. Utilizing a combination of the Diffie-Hellman, RSA, and ElGamal algorithms can offer a strong solution to enhance the security of PDF files against various security threats

Comparison of Deep Neural Network (DNN) and Long Short-Term Memory (LSTM) Algorithms in Email Spam Clustering

2024-03-05T14:16:05+00:00

Sending correspondence through electronic media, also known as e-mail, is a common practice in everyday life. Nonetheless, with the increasing use of e-mail, there is also more misuse of e-mail in the field of correspondence. This research aims to look at the presentation of two algorithms related to email spam clustering, specifically Deep Neural Network (DNN) and Long Short-term Memory (LSTM). The testing process involves determining the model architecture, parameter initialisation. Evaluation is done by comparing accuracy, recall, precision, and F1-score metrics. Testing the performance of DNN and LSTM using Confusion Matrix shows that the DNN algorithm is significantly superior and efficient than the LSTM algorithm. DNN stands out in achieving a higher level of accuracy, where its ability to extract complex features proves more effective. It was found that the DNN algorithm has an accuracy value of 96.32%, recall 96.09%, Precision 96.64% and F1-Score 96.15% while the LSTM algorithm has an accuracy value of 89.20%, recall 87.36%, Precision 90.88% and F1-Score 88.97%.

Classification of Vehicle Types With Deep Learning Based on Image Processing

2024-03-05T14:12:20+00:00

Vehicles play a crucial role in facilitating road traffic and transportation, serving as a primary mode of transportation for people in their daily activities. Various types of vehicles, such as cars, motorcycles, and buses, are commonly utilized for longdistance travel. This study employs Convolutional Neural Network (CNN) and Support Vector Machine (SVM) methods to enhance accuracy in image classification by adjusting the number of epochs and increasing the size of the training dataset. Utilizing a size of 128x128 pixels, the CNN method achieved the highest accuracy rate at 99.33%, surpassing SVM's accuracy rate of 85.06%. Consequently, The Convolutional Neural Network (CNN) method has emerged as the superior choice Compared to the Support Vector Machine (SVM) for image classification tasks.

Support Vector Machine With Feature Selection Chi-Square On Analysis Twitter Sentiment

2024-02-16T20:11:41+00:00

The democracy party that occurred in Indonesia
caused an increase in public comments on social media. Twitter is
one of the most famous social media platforms in Indonesia. By
utilizing a dataset of various public comments on the 2024 general
election, we can do sentiment analysis. Sentiment analysis is
carried out for the purpose of extracting positive or negative
patterns of people's behavior in the implementation of the 2024
election. The algorithm used in analyzing sentiment is a support
vector machine by substantiating chi square in the selection of
dataset features. After testing 2809 data, the results of the
classification accuracy of support vector machine by 73.06%, and
support vector machine with chi square feature selection of
82.77% and F1-score 53.0764 against support vector machine and
F1-score 70.3222 support vector machine with chi square feature
selection.

Identification Of Coffee Fruit Maturity Level Using Machine Learning Based Color Classification With Comparison Of K-nearest Neighbor (K-NN) And Method Support Vector Machine (SVM)

2024-02-16T19:47:14+00:00

Coffee is one of Indonesia's main export commodities which has high economic value. The maturity of coffee berries is an important factor in determining the quality and price of coffee, therefore, developing a method for identifying the level of maturity of coffee berries using image processing is an effective solution. The aim that the author wants to achieve is to obtain a comparison of the performance of K-Nearest Neighbor (KNN) and Support Vector Machine (SVM) to obtain better accuracy values in determining the ripeness of coffee cherries. It was found that for the accuracy value of the image of ripe, quite ripe and raw coffee fruit, namely, the accuracy value obtained using KNN was 98.40%, providing better accuracy compared to SVM which had an accuracy value of 86.90%.

Performance Analysis Algorithm Deep Learning For Introduction Face

2024-02-16T04:46:09+00:00

— Introduction face in world technology own ability
Which Enough Good in do his task. introduction face own
various problem Which can foundlike error position picture face,
part eye, nose as well as ears that are not completely visible and
also with the addition of accessories such as glasses, beards on
picture face Which influence accuracy introduction face.
Algorithm introduction face use Deep Learning with model
network nerve imitation Convolutional Neural Network (CNN).
Results from research that done measure analysis algorithm
Deep Learning with Convolutional Neural Network method for
face recognition use Notation Big-O. With level accuracy
predictions model reached 0.99928075 or about 99.93%. model is
successful identify facial image recognition correctly. Total time
26.18 seconds of execution required to process the image make
predictions with the CNN model. Execution complexity time
algorithm Big-O Notations (O) in introduction image performed
face did not improve significantly with image size (fixed in CNN
model), with constant results of CNN model time complexity as
constant or O(1) time execution recorded around 26 second.
Based on results processtesting training datasets from each of the
two image classes face Rose And Jiso, as much 170 image face
data training, Andvalidation dataset of 80 facial images. Testing
process andmodel execution time results in the level of accuracy
at epoch to 25 val_accuracy as big as 1.00% And total time
execution epochamounting to 151,587 seconds. Which shows that
the algorithm is deep learning method CNN capable identify
introduction face someone with Good.

Application of Natural Language Processing Technology For Making Agricultural Media Chatbots

2024-02-15T08:59:08+00:00

Agriculture is the use of resources by humans to
produce food raw materials and manage the surrounding
environment. With the development of industry today, many fields
use artificial intelligence, including agriculture. The aim of the
research carried out is to help farmers find out some important
information related to agriculture based on current relevance.
Implementation of the application of Natural Language
Processing (NLP) using the Dialogflow Platform to design and
integrate conversational user interfaces for chatbot web
applications. Chatbots have the ability to answer questions and
provide solutions more quickly and efficiently compared to
human-to-human interaction. The contribution of this research is
that Indonesian agriculture is expected to be able to implement
chatbots to improve the quality of information services and for
users to make it easier to obtain information related to Indonesian
agriculture. Based on the results of the evaluation of model
analysis with confusion matrix which was carried out by testing
the reliability method, the results of an accuracy level of 85% were
obtained from the correct number of test data divided by the total
dataset, recall was 100%, precision was 85% and F-measure was
57.14%. This shows that the Chatbot is very feasible and effective
to use.

IoT-Based System for Detecting Grain Moisture Content to Improve Rice Harvest Quality

2024-02-15T08:54:25+00:00

Drying grains serves to remove excess moisture,
prevent rot, and increase shelf life for farmers. The duration of
time required to dry rice is a critical parameter affecting grain
quality and sale value. To achieve ideal grain moisture content,
various farmer groups utilize grain drying machines such as Bed
Dryers. However, drying machines are currently unable to
automatically learn the characteristics of the moisture content of
the grain being dried, necessitating human input to determine the
optimum moisture content. To address this challenge, we propose
the creation of an Internet of Things (IoT)-based grain moisture
content measuring device, which can be integrated into a Bed
Dryer machine. The resulting IoT tool can be augmented with
machine learning, web, or mobile applications. However, this
research is solely focused on IoT tools design and manufacturing.
The ESP32 module functions as a data control and communication
device for each sensor connected to the internet network in IoT
devices. The sensors utilized for measuring grain moisture content
are BME280, Capacitive Soil Moisture, and Negative
Temperature Coefficient (NTC) sensors. The findings present
data on grain temperature values, grain drying environmental
temperatures, and grain moisture content. This information can
serve as a reference point for developing machine learning
algorithms, web applications, and mobile applications that guide
the ideal water content value of grains.

Classification Of Mango Fruit Types Using The Knn Method Based On Shape Feature Extraction

2024-02-15T08:40:31+00:00

—Mango is one of the flagship fruits and is highly
favored by the community. Mangoes are widely loved for their
varied flavors, and with the multitude of mango varieties, it
can be challenging for the general public to classify them. The
types of mangoes commonly consumed by the majority of the
population are harum manis mango, manalagi mango, and
apple mango. Each mango variety has its own distinctive
shape that can be used as a basis in the identification process.
The image capture is done using a mobile phone camera. The
captured images will be used as training data, consisting of 30
images of apple mango, 30 images of harum manis mango, and
30 images of manalagi mango, resulting in a total of 90
training images for various types of mango fruits.
Furthermore, there are 10 test images for each type of apple
mango, harum manis mango, and manalagi mango, resulting
in a total of 30 test images for various types of mango fruits.
Therefore, the total for the entire mango image dataset is 120
digital image data. The K-Nearest Neighbor (KNN) method is
used for the classification of mango types. In this algorithm,
the value of K (number of nearest neighbors) is assumed to be
K=3, using both the Euclidean Distance measurement method
and the Manhattan Distance measurement method. The
method is employed to make a comparison between the
Euclidean Distance measurement model and the Manhattan
Distance measurement model, determining the final
classification results and optimizing the accuracy level in the
KNN algorithm. The results of this research indicate that
using the KNN method with a value of K=3, the Manhattan
distance measurement model is superior to the Euclidean
distance measurement model. The Manhattan distance
measurement model yields an accuracy rate of 91%, Precision
of 91%, Recall of 91%, and F1-Score of 91%.

Comparison of K-NN and Naive Bayes Algorithms to Predict Student Teacher Satisfaction at Madrasah Ibtidaiyah

2024-02-15T08:35:40+00:00

The sentiments that students, parents, the community, or other stakeholders have about the educational process are expressed by their level of satisfaction with education. Reliability, responsiveness, empathy, assurance, and tangible are just a few of the elements that make up a happy student. To predict parental happiness in this study, two approaches were used: K Nearest Neighbors and Naive Bayes. Both Naive Bayes and K Nearest Neighbors demonstrate an accuracy of 0.75 in training comparison findings. On the other hand, KNN outperforms Naive Bayes with an accuracy of 0.69 in the testing comparison.

Predictive Modeling of Graduation Outcomes in Islamic Boarding Schools Using Feedforward Neural Networks

2024-02-14T19:26:20+00:00

Pondok Pesantren Nuur Ar Radhiyyah is an Islamic boarding school that places significant emphasis on Quranic memorization, practical religious practices (ibadah amaliyah), and prayer (ibadah sholat) as core educational values for its students. These values play a pivotal role in determining the graduation status of the santri (students) within the institution. This research investigates the application of feedforward neural networks (FFNN) architecture, focusing on the number of hidden layers and neurons per hidden layer, to develop a predictive model for the future graduation status of the santri. The study explores three different activation functions (Sigmoid, ReLU, and Tanh) and assesses their impact on model performance. Our findings reveal that the optimal model for predicting the santri's graduation status involves the use of the Tanh activation function in combination with four hidden layers and four neurons per hidden layer. This model demonstrates outstanding accuracy, precision, and recall rates of 97.6%.

Development Model Of Higher College Business Incubation Program In The Respati Yogyakarta University

2024-02-15T08:23:17+00:00

UNRIYO, as one of the private universities in Indonesia, has an entrepreneurial vision that targets the application of entrepreneurship in its various activities. In terms of academic activities, UNRIYO has an Entrepreneurship course as a compulsory subject, and other entrepreneurship courses developed by the Faculty such as Technopreneur, Sociopreneurship and Health Technology and Innovation. Based on the results of the 2022 Tracer Study, UNRIYO graduates who have become entrepreneurs have not yet reached the desired number. This is of course UNRIYO's evaluation of the graduate profile which is expected to be in accordance with the entrepreneurial vision. One effort to overcome the problems described above is by increasing collaboration between UNRIYO, industry and the government. Collaboration is carried out through learning activities and activities carried out by UNRIYO through university business units. This collaboration is aimed at the UNRIYO academic community, especially students and academics who have an interest and desire to become entrepreneurs. The university incubation program is one of the efforts that UNRIYO can undertake to achieve its entrepreneurial vision and mission, as well as facilitating the development of startup businesses in Indonesia. This program provides benefits for universities, startup entrepreneurs, government and society. Even though there are still many obstacles faced, with good collaboration between universities, industry and the government, the university incubation program can be an effective solution in developing the creative economy in Indonesia. It is hoped that the UNRIYO higher education incubation program will provide opportunities for UNRIYO to participate in developing Indonesia's creative economy. Start-up businesses that are expected to pass incubation by the UNRIYO Center for Entrepreneurship Studies (ReCEnt) are businesses that apply advances in information technology and have innovation value in their products or methods. This is both a challenge and an opportunity for students to contribute to the development of human civilization in the digital era.

Comparison of Backpropagation with Nguyen Widrow in Heart Attack Classification

2024-02-15T08:15:56+00:00

— Heart disease is a disease that is often fatal and requires
early detection for more effective prevention. Classification of
heart disease through data processing techniques is an important
approach in treating this disease. In this study, heart disease
classification was carried out using two Backpropagation
algorithm methods, namely Backpropagation and
Backpropagation using the Nguyen-Widrow method.
Backpropagation uses random initial weights while weights in
Nguyen-Widrow backpropagation use the Nguyen-Widrow
formula. The comparison results show that the accuracy in
Bacpropagation is 68.3168%, MSE 0.183, Training Time 0.25758
seconds, Precision 0.54455, Recall 1, and F1-Score 0.70513, while
in Bacpropagation on Nguyen-Widrow the accuracy is 66.6667%.
MSE 0.18327, Training Time 0.18736 seconds, Precision 0.54455,
Recall 1, and F1-Score 0.70513.

Face Recognition Using Support Vector Machine (SVM) and Backpropagation Neural Network (BNN) Methods to Identify Gender on Student Identity Cards

2024-02-15T08:12:13+00:00

Until now, many people continue to explore studies on
facial recognition, as reflected in the advancements of Computer
Vision technology implemented in various real-life applications.
This research aims to identify a person's face based on
characteristics or gender features found on student identity cards
at a university. The method employed involves a data science or
machine learning approach, using the SEMMA model (Sample,
Explore, Modify, Model, and Assess) with the application of two
algorithms, namely Support Vector Machine (SVM) and
Backpropagation Neural Network (ANN). This modeling is
further reinforced by pre-processing using Principal Component
Analysis (PCA) to reduce the dimensions of various image
features to selected features. The research results indicate
improved performance, with accuracy reaching 77.50% for the
SVM algorithm and 78.10% for ANN. This performance is
superior to previous studies that did not involve dimension
reduction techniques using PCA.

Implementation of a Plagiarism Detection System Text Based

2024-02-15T08:07:47+00:00

Plagiarism, the act of plagiarizing or stealing work
without acknowledgment, is a serious challenge in the academic
world. Scientific work, as a common target for plagiarism, is
increasingly influenced by information technology. This research
implements a text-based plagiarism detection system by
comparing the level of similarity between the Cosine Similarity
and Jaccard Similarity algorithms against winnowing for text
similarity detection related to variations in N-gram values 3, 5 and
7. Testing was carried out using the Python programming
language and its supporting libraries on 20 datasets sentence. The
test results show that Cosine Similarity is better at detecting
similarities between texts. Accuracy analysis using the confusion
matrix produces an accuracy value of 50%. The comparison
results of different n-gram variations have a total performance
similarity of 15.89% and an average of 0.26%. Meanwhile, the
total performance of Jaccard similarity is 13.59% and the average
is 0.23%. Although Cosine Similarity has higher accu

Comparison of Stroke Classification on Support Vector Machine with Backpropagation Neural Network

2024-02-15T08:03:13+00:00

— Stroke is a serious neurological disease that can cause
long-term health impacts if not detected and treated quickly.
Early detection is crucial for effective prevention and treatment.
In this research, two methods were used, Support Vector Machine
(SVM) and Backpropagation, to predict the possibility of stroke.
The comparison results for Backpropagation training show a
higher accuracy of 0.939 compared to SVM of 0.739, while the
comparison of Backpropagation testing shows a higher accuracy
of 0.879 compared to SVM of 0.787

Comparison of Backpropagation and Backpropagation with Stochastic Gradient Descent

2024-02-15T07:56:33+00:00

Backpropagation has a weakness, namely that
determining the initial weights will affect errors in the training
process, training time, and the accuracy of the resulting model.
Determining the initial weight is difficult to do because the initial
weight determines the error rate, training time and accuracy.
Based on this problem, it is better to optimize weights rather than
looking for ideal weights using stochastic gradient descent which
is expected to reduce errors, reduce training time, and increase
accuracy. This method updates the weights at each
backpropagation iteration by looking at the error value in the
previous iteration. The results of comparing this method are that
backpropagation with stochastic gradient descent has an mse
value of 0.0932, training time 141 seconds, and an accuracy of 80%
compared to backpropagation which has an mse value of 0.1784,
training time 158 seconds, and an accuracy of 64%.
Backpropagation with stochastic gradient descent has been
proven to make backpropagation results better.

Comparative Analysis of Linear Discriminant Algorithms and Support Vector Machine in Palm Fruit Image Disease Classification

2024-02-15T07:51:39+00:00

Artificial Intelligence (AI) is a branch of computer
vision that is used for object detection and image classification
using algorithms. Approaches to comparing object characteristics
in image processing can be divided into High Dimensional Feature
and Low Dimensional Feature approaches. Support Vector
Machine (SVM) is an accurate High Dimensional Feature method,
while Linear Discriminant Analysis (LDA) is a powerful Low
Dimensional Feature method. Some studies combine SVM with
LDA to reduce complexity and improve performance. In the
classification of palm fruit image diseases, the comparison
between SVM (High Dimensional Feature) and LDA (Low
Dimensional Feature) can be done with variations in dataset size
and the percentage of training and testing data in the research is
50:50, 60:40 and 70:30. Performance measurement is based on
accuracy, precision recall and f1-score values. The algorithm for
testing the validity of the accuracy results is k=5 Cross Validation.
The average test result on linear dicriminant analysis, the highest
prediction, namely 0.57, was obtained in the 1st iteration.
Meanwhile, the lowest average value was obtained in the 5th
iteration, namely 0.44. Then the predicted value for system testing
is 0.52. Meanwhile, the calculation of the support vector machine
testing results has the highest average prediction, namely 62.63%,
obtained in the 2nd iteration. The lowest average accuracy value
was obtained in the 4th iteration, namely 62.36%. Then the
prediction value for the testing data system was 62.57, obtained
from the average results of each iteration. This study aims to
compare the performance of the LDA and SVM algorithms in
classifying healthy and diseased oil palm fruit diseases with
variations in dataset size and percentage of training and testing
data. From the results of research conducted, SVM has an
accuracy of 72.02% at the 3rd percentage variation. Meanwhile,
LDA is 69.04% with the same percentage variation. SVM shows
that it is more effective in classifying images of sick palm fruit or
healthy fruit compared to LDA.

Questionnaire Based Hospital Patient Satisfaction Level Classification With Support Vector Machine

2024-02-15T07:47:30+00:00

The utilization of machine learning in various
questionnaire-based classifications, especially using the Support
Vector Machine (SVM) algorithm, has piqued our interest in
conducting research on hospital patient satisfaction levels
through a survey. Using nine questions as features and
measuring the patients' willingness to recommend RS Haji
Medan to others, we built three classification models with
Polynomial, RBF, and Sigmoid kernel functions. Out of the 86
responses we received, our t-test validation test revealed that all
the questions we asked are valid for use in the classification
process. The results show that the Polynomial model produced
the highest accuracy (90.5%), precision (91.8%), and recall
(90.5%) when compared to the RBF and Sigmoid models.
Furthermore, the generated model exhibits stable performance,
with an average difference of less than 7% between the training
and testing performance. This stability suggests promising
resistance to overfitting and underfitting.

Mining PUBG Winning Rates: Between Chicken Dinner and In Game Traveling Distance

2024-02-15T07:43:17+00:00

This research article explores the relationship between
travel distance and player performance in PlayerUnknown's
Battlegrounds (PUBG), a popular battle royale game. Focusing
on walking, riding, and swimming distances, our study employs
machine learning algorithms, including Random Forest (RF),
Support Vector Machine (SVM), and Multilayer Perceptron
(MLP), to predict PUBG player performance, with a primary
emphasis on solo matches. Our research encompasses data
collection, filtering, model configuration, data sampling, and
model evaluation stages to ensure data reliability and model
appropriateness. The results highlight the consistent superiority
of the Random Forest model, which demonstrates the lowest
Mean Squared Error (MSE), Root Mean Squared Error
(RMSE), Mean Absolute Error (MAE), and the highest Rsquared (R2) values in both training and testing phases. In
contrast, the SVM-Sigmoid model consistently delivers poor
predictive performance, emphasizing the significance of selecting
an appropriate model. In summary, this research unveils the
vital role of travel distance in PUBG player performance and
underscores the importance of data-driven decision-making and
model selection in achieving accurate and reliable predictions.
This contribution enhances our understanding of player
performance in the dynamic world of PUBG.

Exploring Student Learning Interests with Support Vector Machine in Pedagogical Strategies

2024-02-15T07:38:30+00:00

In the realm of education and instructional design,
understanding the impact of pedagogical strategies on students'
learning interest is of paramount importance. This study
employs a machine learning approach, specifically Support
Vector Machines with a linear kernel, to comprehensively
explore this relationship. The research investigates the influence
of various pedagogical strategies on students' learning interest
using classification techniques. Our findings reveal robust model
capabilities, with an Area Under the Curve of 93.3%,
Classification Accuracy of 95%, precision of 95.3%, and recall of
95%. Feature importance analysis identifies key contributors,
with the 'Inclusive Learning Environment' and 'Active Class
Discussions' aspects showing significant influence. Our research
underscores the critical role of pedagogical strategies,
particularly in shaping the learning environment and promoting
active class discussions, as they significantly impact students'
learning interest. This study enriches our understanding of the
model's capabilities and highlights the need to consider realworld contexts and validate its performance on external datasets
for successful application and generalization. As educators and
institutions aim to create engaging and effective learning
environments, the insights derived from this research offer
actionable recommendations for improving pedagogical
strategies. This research serves as a valuable resource for those
seeking to enhance the learning experiences of students,
providing a foundation for further exploration of pedagogical
dynamics and their influence on student learning interest.

Enhancing Gender Classification in Higher Education: An Approach with Inception V3 and Backpropagation

2024-02-14T19:32:25+00:00

This research addresses the pressing need for efficient
and accurate gender classification of college applicants within
academic information systems. Current methods involve manual
gender selection, often leading to inaccuracies. We present a deep
learning model that automatically classifies gender using
passport-sized images, leveraging advanced techniques. Our
approach streamlines the admissions process, enhancing data
accuracy, and contributing to gender classification in educational
settings. In the realm of deep learning, we explore integrating
Inception V3 with Backpropagation, using features extracted
from Inception V3 for classification. We collected 160 balanced
training images and an unbiased validation dataset to ensure
real-world applicability. Results from our models (NN01-NN09)
demonstrate impressive accuracy, precision, and recall. NN02
consistently excels across all metrics, making it an ideal choice
for practical deployment. Validation results suggest room for
improvement in handling diverse data sources. In conclusion, our research improves gender classification in higher education,
emphasizing the value of modern technology. NN02 is
recommended for real-world applications, emphasizing the
significance of efficient gender classification in improving the
college applicant experience.