IKIchwanul Muslim Karo Karo

Ichwanul Muslim Karo Karo

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Ichwanul Muslim Karo Karo has a total of 29 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    An optimized kernel SVM framework for game review sentiment analysis using particle swarm optimization

    by Ichwanul Muslim Karo Karo, Sri Dewi, Alvi Sahrin Nasution

    The digital game industry’s rapid growth has increased interaction between developers and players through online review platforms. These reviews contain vital information about gaming experiences and satisfaction, serving as guides for future game versioning. Sentiment analysis provides a strategic approach to automatically classify reviews, offering data-driven insights for developers. This study focuses on enhancing sentiment analysis performance for Player Unknown's Battlegrounds (PUBG) reviews by integrating Kernel Support Vector Machine (SVM) with Particle Swarm Optimization (PSO). A dataset of 1,205 reviews from the Google Play Store was analyzed using TF-IDF feature extraction and 5-fold cross-validation. While default Kernel SVM achieved 76.78% accuracy, it suffered from low precision (56.9%). Implementing PSO for parameter optimization significantly improved performance, reaching 86.42% accuracy, 70.98% precision, and an 83.03% F1-score. Comparisons with Naïve Bayes, basic SVM, BERT, and Lexicon + SVM confirm that the Kernel SVM + PSO model provides superior and more stable performance. These findings highlight PSO’s effectiveness in SVM parameter tuning. Future research should investigate combining metaheuristic optimization with deep learning models to improve model generalization.

  • Open Access

    Indonesian Elementary Students’ Perceptions of Teachers’ Affective Support: A Cluster Analysis Using National Literacy and Numeracy Assessment Data

    by Rizki Habibi, Muliawan Firdaus, Ichwanul Muslim Karo Karo

    Affective support from teachers—such as academic expectations, attention and care, and constructive feedback—plays a critical role in students’ learning outcomes but is often underexplored in large-scale educational assessments, particularly in developing countries. This study examines how Indonesian elementary students perceive teacher affective support and how these perceptions relate to their literacy and numeracy performance. Using data from the 2023 Indonesian National Assessment involving 214,481 fifth-grade students, we employed K-Means clustering to identify latent student profiles based on their literacy, numeracy, and self-reported perceptions of teacher support. Variables were normalized, and the optimal number of clusters was determined using the Elbow, Silhouette, and Davies-Bouldin methods. Five distinct student clusters emerged, each characterized by unique combinations of academic achievement and affective perceptions. High-achieving students consistently reported more positive perceptions of teacher support, particularly in terms of feedback and expectations. ANOVA tests confirmed significant differences (p 0.001) across clusters in all affective and academic variables, with moderate to large effect sizes. The findings highlight the alignment between academic success and perceived teacher affective support. This clustering approach reveals nuanced student profiles that traditional methods may overlook, offering a data-driven foundation for differentiated teaching, teacher training, and policy interventions. Clustering national assessment data provides actionable insights for enhancing affective support in classrooms. The methodology is scalable and adaptable for use in other educational systems seeking to personalize instruction and promote equity.

  • Open Access

    Hair Disease Classification Using Convolutional Neural Network (CNN) Algorithm with VGG-16 Architecture

    by Ichwanul Muslim Karo Karo, Dedi Kiswanto, Suvriadi Panggabean, Adidtya Perdana

    Hair diseases are common and can be caused by a variety of factors, including genetics, stress, nutritional deficiencies, as well as exposure to sunlight and air pollution. Accurate diagnosis of hair diseases is important for proper treatment, but can be challenging due to overlapping symptoms. The development of the healthcare world has widely utilized machine learning and deep learning approaches to assist in the healthcare field. This research aims to develop hair disease classification using Convolutional neural network (CNN). The CNN-based approach is expected to help health professionals diagnose hair diseases accurately and provide targeted treatment. This research involves an experimental design with three main stages: identifying the research problem, conducting a literature review, and collecting data. The research uses a dataset of hair disease images obtained from Kaggle, which are annotated and organized based on different hair disease types. After the image data is collected, the image dataset will go through the image preprocessing stage. Experiments were conducted using hair disease image data with 15 epochs on a CNN Deep Learning model with VGG-16 architecture, and resulted in an accuracy of 94.5% and a loss rate of 18.47%, with a testing epoch time of 9 hours 48 minutes. The results of this study show that CNN with VGG-16 architecture can successfully classify 10 types of hair diseases


  • Open Access

    Comparison of Activation Functions on Convolutional Neural Networks (CNN) to Identify Mung Bean Quality

    by Ichwanul Muslim Karo Karo, Justaman Arifin Karo Karo, Manan Ginting, Yunianto Yunianto, Hariyanto Hariyanto, Novia Nelza, Maulidna Maulidna

    Mung bean production levels by farmers in Indonesia are not stable. When there is a surplus, the stock of mung beans in the warehouse will accumulate, the storage factor affects the quality of mung beans. Indicators of quality mung beans can be seen from the color and size through direct observation. However, the aspect of view and assessment and the level of health of each observer is a human error in the classification of mung bean quality so that the results are less than optimal. One alternative way to identify object quality is to use deep learning algorithms. One of the popular deep learning algorithms is convolution neural network (CNN). This study aims to build a model to classify the feasibility of mung beans. The process of building the model also goes through the image preprocessing stage. In the process of building the model, there are ten setup parameters and four setup data used to produce the best model. As a result, the best CNN algorithm model performance is obtained from data setup I, with accuracy, precision, recall and F1 score above 75%. In addition, this study also analyzes Rel U and Adam activation functions on CNN algorithm on model performance in identifying mung bean quality. CNN algorithm with Adam activation function has 92% accuracy, 92.53% precision, 91.9% recall, and 92.19% F1 score. In addition, the performance of CNN algorithm with Adam activation function is superior compared to CNN algorithm with Adam activation function and previous study


  • Redesign of Indonesia E-commerce Online Review using User Information Behavior and Build-Learn-Measure Approaches

    by Dana Sulistiyo Kusumo, Dita Noviyanti, Arfive Gandhi, Ichwanul Muslim Karo Karo, Su-Cheng Haw, Palanichamy Naveen

    Online review is a critical feature in e-commerce to increase users' decisions to buy products. This research found that an e-commerce platform in Indonesia lacked design aspects in its online review and recommender system results that should be improved. This research combined user information behavior (UIB) with a lean startup approach (build-measure-learn). To achieve the expected user goals, designing an online review in e-commerce must consider how users interact with the online review and recommender system. It also should consider the e-commerce users' and owners' perspectives. This research improved and evaluated the design of the existing e-commerce in Indonesia. Firstly, it adopted the UIB approach to detail the design of the e-commerce online review and recommender system's results related to the review features. Then, it performed the initial usability testing to get rapid feedback from ten study participants using the maze tools to measure a usability task and heatmap tracing. Lastly, this research followed up and evaluated the results of the initial usability testing using the build-measure-learn approach. The results revealed that UIB implementation for designing the online review and recommender system results as follows. The redesign had better misclick rate (31.7% compared to original design 41.7%), better average success (46.2% compared to original design 33.3%), easier to use (4.3 in 5 likert scale compared to original design 3.7) and all participants chosen the multiple filter feature compared to single filter in original design and participants chosen to the addition of like and dislike feature in the proposed design. These results were captured from the participants study through rapid feedback based on the user information behavior when using the online review. From a design perspective, it can help an e-commerce designer to improve other aspects of the e-commerce features.


  • Open Access

    Movie recommender chatbot based on Dialogflow

    by Zinke Abdurahman Baizal, Nurul Ikhsan, Ichwanul Muslim Karo Karo, Reinaldo Kenneth Darmawan, Roby Dwi Hartanto

    Currently, the online movie streaming business is growing rapidly, such as Netflix, Disney+, Amazon Prime Video, HBO, and Apple TV. The recommender system helps customers in getting information about movies that are in accordance with their wishes. Meanwhile, the development of messaging platform technology has made it easier for many people to communicate instantly. Utilizing a messaging platform to build a recommender system for movies, provides special benefits because people often access the messaging platform all the time. In the Indonesian language, there are many slang terms that the system must recognize. In this study, we build a chatbot on a messaging platform which users can interact with the system in natural language (in Indonesian language) and get recommendations. We use rule-based and maximum likelihood as a method in natural language processing (NLP), and content-based filtering for the recommendation process. The recommender system interaction is built through a conversation mechanism that will form a conversational recommender system. The interaction is based on a chatbot which is built using Dialogflow and implemented on the telegram. We use the accuracy of recommendations and user satisfaction to evaluate the system performance. The results obtained from the user study indicate that the NLP approach provides a positive experience for users. In addition, the system also produces an accuracy value of 83%.


  • Open Access

    Pelatihan media e-learning classroom untuk guru SMKN 1 Peureulak Timur

    by Ichwanul Muslim Karo Karo, Widi Astuti, Ramanti Dharayani

    One of the sectors affected by the covid-19 pandemic is the education sector. SMKN 1 Peureulak Timur as the only vocational high school in Peureulak Timur sub-district, East Aceh must continue to provide education services for all students of the best quality. One of the efforts to improve educating and teaching skills for teachers at SMKN 1 Peureulak Timur is by providing e-learning training. One of the e-learning media that is often used is Goggle Classroom because of its complete features and quite easy to operate. Therefore, the activity that will be carried out at this community dedication is Google Classroom training. It is hoped that after receiving this Google Classroom training, teachers of SMKN 1 Peureulak Timur can improve their performance in facing the Industrial Revolution 4.0.


  • Open Access

    Analysis of Expertise Group Using The Fuzzy K-NN Classification Algorithm (Case Study: School of Computing Telkom University)

    by Jodi Kusuma, Angelina Prima Kurniati, Ichwanul Muslim Karo Karo

    The School of Computing at Telkom University has four Expertise Groups that defines the lectures taken by students. Deciding the Expertise Group, will be influential in deciding elective courses and raising the topic of the Final Project. There are many students who are still having difficulty in deciding the Expertise Group and finally only decide based on the most popular Expertise Group without seeing their potential and abilities. The impact of wrong decision of the Expertise Group are delays in graduation time. It will then affect accreditation of study program and university rank, especially in the timely graduation indicator. Therefore, it is necessary to have a system that can predict the decision of the Expertise Group for the School of Computing students based on their academic scores. In this study, prediction using the Fuzzy K-Nearest Neighbor classification algorithm was chosen because it can determine the class based on the nearest neighbor and consider ambiguous data because of the weighting value in each class. There are five tests carried out to get the best model, namely (1) examine the best split training and validation data, (2) examine the best K value, (3) compare Fuzzy K-Nearest Neighbor with Naïve Bayes and Decision Tree (C4.5) which is a commonly used classification algorithm, (4) examine the values of accuracy, precision, recall, f1-score, and (5) examine the values of accuracy using Cross-Validation method. The result is that the model made using Fuzzy K-Nearest Neighbor has an accuracy value of 72% in the case of imbalance data, 62% in the case of applying the undersampling technique, and 56% in the case of applying oversampling. Based on experiments with the other two algorithms, it was found that compared to the other two algorithms, the Fuzzy K-Nearest Neighbor has a higher accuracy value in the case of imbalance data and the case of applying to undersampling, but it has a lower accuracy in the case of applying oversampling, due to the lack of Fuzzy K-Nearest Neighbor in handling small minority data variations.

  • Open Access

    Pengaruh Metode Pengukuran Jarak pada Algoritma k-NN untuk Klasifikasi Kebakaran Hutan dan Lahan

    by Ichwanul Muslim Karo Karo, Ananda Khosuri, Juan Steiven Imanuel Septory, Dimas Pebrian Supandi

    Forest and land fires are a serious and recurring problem in Indonesia. The high intensity of forest fires is caused by the distribution of hotspots in fire-prone areas. One of the efforts to prevent and minimize the risk of forest fires is to identify the types of hotspots using a classification approach. One of the most popular classification algorithms is k Nearest Neighbor (k-NN). The algorithm uses a distance calculation approach in classifying objects. The purpose of this study is to classify the types of hotspots scattered in Indonesia using the k-NN algorithm and to analyze the effect of the distance calculation method on the k-NN algorithm. The types of distance measurement methods analyzed include Euclidean, Canberra, Chebyshev, and Manhattan. The dataset used is the distribution of hotspots in Indonesia obtained from Global Forest Watch (GFW). The study designed a dataset with two conditions, through the pre-processing stage and not. In general, the model accuracy of the k-NN combination with various distance measurement methods is above 90%. The pre-processing stage can increase the model's performance 1-8 times. The combination of k-NN with Manhattan is the best choice to identify the types of hotspots with an accuracy of 92.6%.