Jurnal Ilmu Komputer dan Informasi

An alternative for kernel SVM when stacked with a neural network

2024-04-23T09:22:20+07:00

Many studies stack SVM and neural network by utilzing SVM as an output layer of the neural network. However, those studies use kernel before the SVM which is unnecessary. In this study, we proposed an alternative to kernel SVM and proved why kernel is unnecessary when the SVM is stacked on top of neural network. The experiments is done on Dublin City LiDAR data. In this study, we stack PointNet and SVM but instead of using kernel, we simply utilize the last hidden layer of the PointNet. As an alternative to the SVM kernel, this study performs dimension expansion by increasing the number of neurons in the last hidden layer. We proved that expanding the dimension by increasing the number of neurons in the last hidden layer can increase the F-Measure score and it performs better than RBF kernel both in term of F-Measure score and computation time.

Improving Classification Performance on Imbalanced Medical Data using Generative Adversarial Network

2024-04-23T09:18:04+07:00

In many real-world applications, the problem of data imbalance is a common challenge that significantly affects the performance of machine learning algorithms. Data imbalance means each target of classes is not balanced. This problem often appears in medical data, where the positive cases of a disease or condition are much fewer than the negative cases. In this paper, we propose to explore the oversampling-based Generative Adversarial Networks (GAN) method to improve the performance of the classification algorithm over imbalanced medical datasets. We expect that GAN will be able to learn the actual data distribution and generate synthetic samples that are similar to the original ones. We evaluate our proposed methods on several metrics: Recall, Precision, F1 score, AUC score, and FP rate. These metrics measure the ability of the classifier to correctly identify the minority class and reduce the false positives and false negatives. Our experimental results show that the application of GAN performs better than other methods in several metrics across datasets and can be used as an alternative method to improve the performance of the classification model on imbalanced medical data.

Note on Algorithmic Investigations of Juosan Puzzles

2024-04-23T09:12:27+07:00

We investigate several algorithmic and mathematical aspects of the Juosan puzzle—a one-player pencil-and- paper puzzle introduced in 2014 and proven NP-complete in 2018. We introduce an optimized backtracking technique for solving this puzzle by considering some invalid subgrid configurations and show that this algorithm can solve an arbitrary Juosan instance of size m × n in O(2^mn) time. A C++ implementation of this algorithm successfully found the solution to all Juosan instances with no more than 300 cells in less than 15 seconds. We also discuss the special cases of Juosan puzzles of size m × n where either m or n is less than 3. We show that these types of puzzles are solvable in linear time in terms of the puzzle size and establish the upper bound for the number of solutions to the Juosan puzzle of size 1 × n. Finally, we prove the tractability of arbitrary m × n Juosan puzzles whose all territories do not have constraint numbers.

Improving Remote Sensing Change Detection Via Locality Induction on Feed-forward Vision Transformer

2024-04-23T09:09:46+07:00

The main objective of Change Detection (CD) is to gather change information from bi-temporal remote sensing images. The recent development of the CD method makes use of the recently proposed Vision Transformer (ViT) backbone. Despite ViT being superior to Convolutional Neural Networks (CNN) at modeling long-range dependencies, ViT lacks a locality mechanism, a critical property of pixels that comprise natural images, including remote sensing images. This issue leads to segmentation artifacts such as imperfect changed region boundaries on the predicted change map. To address this problem, we propose LocalCD, a novel CD method that imposes the locality mechanism into the Transformer encoder. Particularly, it replaces the Transformer's feed-forward network using an efficient depth-wise convolution between two $1 \times 1$ convolutions. LocalCD outperforms ChangeFormer by a significant margin. Specifically, it achieves an F1-score of 0.9548 and 0.9243 on CDD and LEVIR-CD datasets.

Implementation Genetic Algorithm for Optimization of Kotlin Software Unit Test Case Generator

2024-04-22T15:22:24+07:00

Unit testing has a significant role in software development and its impacts depend on the quality of test cases and test data used. To reduce time and effort, unit test generator systems can help automatically generate test cases and test data. However, there is currently no unit test generator for Kotlin programming language even though this language is popularly used for android application developments. In this study, we propose and develop a test generator system that utilizes genetic algorithm (GA) and ANTLR4 parser. GA is used to obtain the most optimal test cases and data for a given Kotlin code. ANTLR4 parser is used to optimize the mutation process in GA so that the mutation process is not totally random. Our model results showed that the average value of code coverage in generated unit tests against instruction coverage is 95.64%, with branch coverage of 76.19% and line coverage of 96.87%. In addition, only two out of eight generated classes produced duplicate test cases with a maximum of one duplication in each class. Therefore, it can be concluded that our optimization with GA on the unit test generator is able to produce unit tests with high code coverage and low duplication.

A Dynamic-Bayesian-Network-Based Approach to Predict Immediate Future Action of an Intelligent Agent

2024-04-22T15:18:58+07:00

Predicting immediate future actions taken by an intelligent agent is considered an essential problem in
human-autonomy teaming (HAT) in many fields, such as industries and transportation, particularly to
improve human comprehension of the agent as their non-human counterpart. Moreover, the results of such
predictions can shorten the human response time to gain control back from their non-human counterpart
when it is required. An example case of HAT that can be benefitted from the action predictor is partially
automated driving with the autopilot agent as the intelligent agent. Hence, this research aims to develop an
approach to predict the immediate future actions of an intelligent agent with partially automated driving
as the experimental case. The proposed approach relies on a machine learning method called naive Bayes
to develop an action classifier, and the Dynamic Bayesian Network (DBN) as the action predictor. The
autonomous driving simulation software called Carla is used for the simulation. The results show that the
proposed approach is applicable to predict an intelligent agent’s three-second time-window immediate future
action.

Enhancing Assault Maneuvers in Simulated Scenarios of Multiple Invader Kamikaze Drones through the Utilization of a Modified Adaptive Elforce Algorithm

2024-04-22T14:10:01+07:00

The development of autonomous drone technology has led in their widespread deployment,
especially in combat scenarios. One instance of this is the utilization of kamikaze drones, as
seen in the Ukraine war. Autonomous defense drones have been used to counter these
invading kamikaze drones. This study focuses on simulating scenarios involving invader vs.
defender drones, primarily exploring invader drone maneuver motions to maximize damage
inflicted on chosen targets. The work we conducted presents an enhanced el-force algorithm
that employs Coulomb's Law-based maneuver techniques to improve the effectiveness of
multiple kamikaze invader drones when engaging target defended by defender drones. We
aim to improve traditional el-force by addressing key challenges such as siege tendencies and
unproductive conduct. In addition, we explore various attacking formations to determine the
most effective formation. To evaluate the performance of our proposed algorithm, we
conducted simulation in a dynamic 3D environment, employing damage inflicted as the
evaluation metric. Through rigorous testing, we conclusively demonstrate that our proposed
method combining with a circular formation, outperforms alternative attacking maneuvers
and formations. Our findings provide insights into optimal maneuver movements and
attacking formations, improving the effectiveness of invader drones in engaging and
damaging designated targets.

Predicting Earthquake Magnitudes in Indonesia: Exploring the Potential of the Prophet Algorithm

2024-04-22T14:03:34+07:00

Research on earthquakes has been extensively conducted by previous studies using various methods and specific discussions. Similarly, research to predict the magnitude of earthquakes that will occur in the future has also been conducted. This study employs the Prophet algorithm to test its capability in predicting a case study's magnitude using data with numerous missing values and outliers. The study is conducted without transformation and with Box-Cox and log-transformations. Transformations are applied to handle outliers. The results indicate that across the three experiments, the difference between the predicted and actual data ranges from 0.1 to 0.5 or even more. Performance metrics reveal that the log-transform is superior to the other two experiments, with a smaller MAE of 0.27 and a MAPE of 5.96%. Nevertheless, the use of the Prophet algorithm in this case study needs further investigation with different treatments to achieve more accurate results.

Land Cover Segmentation of Multispectral Images Using U-Net and DeeplabV3+ Architecture

2024-04-22T13:54:09+07:00

The application of Deep Learning has now extended to various fields, including land cover classification. Land cover classification is highly beneficial for urban planning. However, the current methods heavily rely on statistical-based applications, and generating land cover classifications requires advanced skills due to their manual nature. It takes several hours to produce a classification for a province-level area. Therefore, this research proposes the application of semantic segmentation using Deep Learning techniques, specifically U-Net and DeepLabV3+, to achieve fast land cover segmentation. This research utilizes two scenarios, namely scenario 1 with three land classes, including urban, vegetation, and water, and scenario 2 with five land classes, including agriculture, wetland, urban, forest, and water. Experimental results demonstrate that DeepLabV3+ outperforms U-Net in terms of both speed and accuracy. As a test case, Landsat satellite images were used for the Karawang and Bekasi Regency areas.

Rethinking Smart Keyboard Layout to Aid Strong Password Creation

2024-04-22T09:49:51+07:00

In an era marked by increasing digitization and the omnipresence of smartphones, the importance of robust
password security cannot be overstated. With the ever-growing threat of cyberattacks, there is a pressing need
for user-friendly tools that facilitate the creation of strong and unique passwords. Traditional alphanumeric
keyboard layouts (physical or virtual) have remained largely unchanged for decades, relying on the same
QWERTY layout initially designed for typewriters. However, these layouts may not be optimal for generating
strong passwords. This paper focuses on tailoring virtual keyboard layouts on smartphones specifically for
strong password creation. For this, we have performed extensive user surveys to see if the presence of
dedicated rows for digits and special characters (essential in any strong password) allows users to create
stronger passwords compared to regular smartphone keyboard layout. Apart from that, we also investigated
the optimal assignment of characters, digits, and special characters and their groupings in a single soft key.
The findings from the detailed user experiment suggested optimal settings for a smartphone virtual keyboard
(for Android) like- diagonal length for good typing speed (approximately between 8.38 and 9.41 cm), and
key density (0.88 to 1.21 keys/cm2) which produces the least error without sacrificing the strength of
passwords created using those layouts. We hope the outcome of this paper will help designers to aid virtual
keyboard layouts for smartphones that can motivate and create strong passwords without sacrificing usability.