Research in SoCS focuses in Artificial Intelligence and Cybersecurity. We are interested in developing concepts in Machine Learning, Neural Network, Deep Learning, and Pattern Recognition which can be used in Data Science especially for data analysis. As for CyberSecurity, we are focusing on Network Penetration Testing (Vulnerability Assessment), Information Security Management (cyber threat intelligence/CTI), Cryptography and its Applications, and Threat Analysis/Security Analysis (Malware Analysis & Reverse Engineering).

Artificial Intelligence

Visualization of Interactive Navigation Paths in Forest-Mountain and Urban environments
Determining Shortest Path for Disaster Emergency Route with Augmented Reality Environment Using A-Star and Dijkstra Algorithms

Title:

Determining Shortest Path for Disaster Emergency Route with Augmented Reality Environment Using A-Star and Dijkstra Algorithms

Author:
Zaenal Alamsyah

Supervisor:
1. Prof. Media Anugrah Ayu
2. Prof. Teddy Mantoro

Abstract:

A natural disaster is a condition of an event or activity in nature that causes several negative impacts such as loss, damage, suffering as well as psychological disturbances in individuals. These natural disasters can occur due to nature itself or due to human activity. Starting from 1 January – until 31 December 2020, victims of natural disasters in Indonesia reached 6,450,903 people and total disaster is 2.952 cases. Due to the frequent occurrence of natural disasters in Indonesia, disaster management is needed. One of the problems in disaster management is finding the nearest disaster emergency route, studies that focus on route search navigation are still very rare. Augmented Reality is a technology that is being developed to help visualize an object in the form of images or 3D objects that help provide real information. This study focuses on route search navigation in Augmented Reality using the A-star, Dijkstra for comparison. The purpose of this research is to navigation the path search for disaster emergency routes in augmented reality so as to provide direct visualization to mark the path. For this reason, using several algorithms to be compared so that the path search navigation in augmented reality can be done and produces the best.

Keywords – Shortest Path, Augmented Reality, A-Star, Dijkstra

DEMO: CLICK HERE

Title:

Navigation on a 3D Mountain Map Using A-Star Algorithm for Searching Missing Persons

Author:
Muhamad Ikhsan Thohir

Supervisor:
1. Prof. Media Anugrah Ayu
2. Prof. Teddy Mantoro

Abstract:

Indonesia is a country that has a lot of forests, especially mountains, the number of cases of people missing when climbing is due to a lack of experience and competence during emergencies and climber risk management. Finding missing persons in remote areas, especially mountains, is often a challenge for rescuers because of the lack of terrain guidance in the area. Terrain that is dangerous for rescuers while exploring the area makes the rescue time even longer. To be able to solve this problem 3D maps can be used to visualize precise information to assist volunteers in searching for missing people in the mountains. The Geographic Information System (GIS) and the A-Star algorithm are combined to find the closest path to the location of the missing person. The results show that the A-star can be combined with GIS and visualized well using a 3D map.

Keywords— A-Star, GIS, 3D Map, Missing Person.

 

Crime Detection
Crime Rate Detection Based on Text Mining on Social Media Using Multi-model Algorithm

Tittle:

Crime Rate Detection Based on Text Mining on Social Media Using Multi-model Algorithm
Author :

M. Anton Permana
Supervisor :

Prof. Ir.Teddy Mantoro.,MSc.,PhD.,SMIEEE & Prof.Ir. Media Ayu.,MSc,PhD,SMIEEE

Abstract: 

Social media recently are very populer in Indonesia and in the world. Luckly, this platform may express their opinion and emotion even other party especially reseacher mostly use this opportunity to find any solution for any case likely competitive business, decisions maker, and possible analitics and predictive support system. In this case our analysis is content on Twitter and Facebook which user often posted information about the crime which matters need police attention. Therefore our purpose is detection of crime rate on social media to find pattern trend of tweet number crime. This work used text mining approach for classification of tweet and post content text into 10 class of crime. The Algorithm used for classifier are Logistic Reggression, Naïve Buyes, Support Vector Machine (SVM) and Decission Tree. from all the algorithm used, Logistic regression give the best accuracy of 90%.

Keywords: Crime Rate Detection, Text Mining, Classification, Social Media

Face Detection
Face Mask Detection in The Covid 19 Pandemi Era by Implementing Convolutional Neural Network and Pretrained CNN Model

Tittle:
Face Mask Detection in The Covid 19 Pandemi Era by Implementing Convolutional Neural Network and Pretrained CNN Model

Author:
Ivana Lucia Kharisma

Supervisor:
Rahmadya Trias Handayanto ST, MKom, PhD
Deshinta Arrova Dewi S.Kom , M.Si, PhD      

Abstract:
The Coronavirus or Covid-19 has spread widely throughout the world since the beginning of 2020. The virus that originated in the Wuhan area, China has been declared a pandemic worldwide. Believed to be contagious due to interaction between humans, at this time this virus has infected more than 200 countries around the world, with the number infected exceeding 20 million people. With the massive development of this virus, various methods in the medical field, ranging from prevention, disease detection, and treatment measures are developed to reduce the rate of positive increase of Covid-19. WHO provides basic guidance in forbidding the spread of the virus that can be done by the community.  One of them is the use of masks when doing activities outside the home. The objective of this research is to develop a face mask detection model based on the current state of the art for face mask classification model. The research methodology consists of three major phases, which are analysis of the Problems of face mask detection using a convolutional neural network, implementing the pre-trained convolutional neural network model for develop the face mask detection, and doing analysis of the performance from the models developed. In the training process, the accuracy of CNN, VGG16, and VGG19 are  97.9%, 99.87% and 100%,  respectively. In model evaluation using classification reports, CNN model provides the best value in calculating accuracy , precision, recall and F1-score for all classes, masks and numbers compared to model using pretrained CNN models .

Keywords—Covid19, Face Mask, Convolutional Neural Network

Demo: CLICK HERE

Deep Learning
Fabric Defect Detection with Deep Learning Algorithm for Textile Quality Control

Tittle:
Fabric Defect Detection with Deep Learning Algorithm for Textile Quality Control

Author:
Kamdan

Supervisor:
Dr. Dini Oktarina Dwi Handayani
Dr. Rahmadya Trias Handayanto 

Abstract:

Quality control is a major aspect of manufacturing, especially the fabric industry. In the textile industry, good tools are needed to help increase productivity and profits. For this reason, a good method of monitoring and detection of fabric defects is needed. The proposed method designed using VGG19, CNN and VGG16 algorithm. This study aims to create a defect detection model in the production of fabric making. After conducting research with a quantitative approach and training with the proposed method, it turns out that the detection accuracy of fabric defects has increased. In the training process, the accuracy using VGG19, CNN and VGG16 using 10 epochs are 0.9712, 0.5150 and 0.6237 respectively. Using 100 epochs, the accuracy using VGG19, CNN and VGG16 are 1.000, 0.9825 and 0.7337. By using 0.001 learning rate, the accuracy given from VGG19, CNN and VGG16 are 0.7987, 0.5138 and 0.5150. Using the 2×2 kernel size, the accuracy given from VGG19, CNN and VGG16 models are 0.9650, 0.4925 and 0.6212. For all the training process by change some parameters for comparison, VGG19 gives better accuracy than modelling using CNN and VGG16.

Keywords: Deep Learning, CNN, VGG16, Textile

Deep Learning Approaches to Identify Sukabumi Potentials Through Images on Instagram

Tittle:

Deep Learning Approaches to Identify Sukabumi Potentials Through Images on Instagram

Author:

Dede Sukmawan

Supervisor:

Dr. Dini Oktarina Dwi Handayani

Dr. Deshinta Arrova Dewi

Abstract:

Sukabumi Regency is one of the largest regencies on the island of Java. With a large area and a fairly dense population, it creates its own problems, such as in managing the potential of places and communities. The purpose of this research is to explore the potential of Sukabumi Regency through Instagram social media with #sukabumiupdate. Data collection is done by taking pictures from social media Instagram, the data taken is 6,970 images. Each data that has been collected is divided into 4 (four) class categories based on the type of image, namely Tourism class, culture class, culinary class, and handicrafts. Then the data is classified using a deep learning approach with three methods, namely CNN, VGG16, and VGG19. These three models are very good at image processing. From the results of data processing through the CNN approach, the accuracy value is up to 91%, then the VGG16 approach has an accuracy value of 99%, and finally, through the VGG19 approach, the accuracy is 95%. So it can be ascertained that from the three models of the deep learning approach the best accuracy value is VGG16.

Keywords:

Deep Learning, CNN, VGG19, Social Media, Instagram

Indoor Tracking User Location
Determining Indoor User Location Based on Signal Strength of IEEE 802.11 Using Machine Learning Techniques

Tittle:

Determining Indoor User Location Based on Signal Strength of IEEE 802.11 Using Machine Learning Techniques   

Author:
Gina Purnama Insany

Supervisor:
Prof. Ir. Media Anugerah Ayu, M.Sc., PhD., SMIEEE
Prof. Ir. Teddy Mantoro, M.Sc., PhD., SMIEEE

Abstract:

GPS provides enormous benefits to the navigation system, unfortunately this GPS technology has low accuracy when the user is indoors due to limited signals that cannot be reached. For that a more accurate system ineeded, one of which is the Indoor Positioning System (IPS). The concept of indoor localization uses a lot of Wireless Local Area Network (WLAN)/IEEE 802.11 technology because WLAN is almost available anywhere and can be easily integrated with a smartphone. WLAN have been widely used to facilitate communication over Wi-Fi networks. By collecting signal strength (Receive Signal Strength/RSS) data from several different Access Points (APs), a person’s position in the building can be calculated and determined. RSS measurement is done using Wi-Fi Netgear and data processing is done using by Google Colab. The training data and testing data are processed using the machine learning techniques such as        k-Nearest Neighbour (k-NN), Naïve Bayes, Decision Tree, and SVM algorithm. The implementation of results with the WLANs method are expected to obtain location accuracy values ​​for indoor user locations. The k-NN_3 and k-NN_5  have the optimum accuracy  (91%) and the smallest error rate (0.1875) among other algorithms.

Keywords: IPS, WLAN/IEEE 802.11, Machine Learning Techniques, Accuracy Values.

Demo: Click Here

Cyber Security

Malware Detection
Malware Detection and Classification Model Using Machine Learning Random Forest Approach

Tittle:

Malware Detection and Classification Mo del Using Machine Learning Random Forest Approach

Author:

Carti Irawan

Supervisor:

Prof. Ir. Teddy Mantoro, M.Sc., PhD., SMIEEE

Prof. Ir. Media Anugerah Ayu, M.Sc., PhD., SMIEEE

Abstract:

Malware programs attack computer systems, smart mobile devices, and some applications. Malware is a program that needs to be watched out for because it can be a threat to computer users and internet networks. Malware was created to steal personal information about computer users or control user devices over a network. Computers are easily infiltrated by various malware programs that can interfere with and even damage user files. To solve this problem, this study discusses malware detection based on network traffic and classifies the types of malware based on their groups in Personal Computer so that it can help detect these types of malware and classify using Machine Learning method with Random Forest Approach Algorithm compared to the Decission Tree and Gradient Boosted Tree for the Accuracy. Random Forest has the highest level of accuracy. Data used in this study were is taken from the Kaggle Data Collection on Android Network Traffic Malware.

Keywords: Malware, Machine Learning, Random Forest, Malware Analysis, Smart Mobile Device