AFAhmad Fauzi

Ahmad Fauzi

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

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Ahmad Fauzi has a total of 60 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Transfer learning-based malnutrition classification using VGG16 and comparative analysis of CNN architectures

    by Ahmad Fauzi, Haerul Yuda Aditiya, Maharina Maharina, Sihabudin Sihabidin, Muhammad Ansari Adista, Iflan Naufal, Natasya Eka Nanda Sonia Puri, Candra Zonyfar

    Childhood malnutrition remains a critical public health challenge in developing countries, with Indonesia ranking fifth globally for stunting burden. Traditional anthropometric assessment methods are time-consuming, resource-intensive, and require trained personnel, necessitating efficient computer-based early detection approaches. This study proposes a deep learning-based method for automated nutritional status classification using facial image analysis. We developed and compared multiple transfer learning architectures including visual geometry group 16 (VGG16), densely connected convolutional network 121 (DenseNet121), mobile network version 2 (MobileNetV2), and residual network 50 (ResNet50) for classifying children into three categories: healthy, malnutrition, and stunting. Results demonstrated that VGG16, a simpler architecture trained for only 10 epochs, achieved optimal performance with 91.8% accuracy, and significantly outperforming more complex modern architectures like ResNet50. This finding challenges the conventional assumption that newer, deeper models invariably perform better, and particularly when working with limited medical datasets. The study revealed that longer training durations led to performance degradation due to overfitting, emphasizing the importance of balancing model complexity with dataset characteristics. These findings support the development of practical artificial intelligence (AI)-based malnutrition screening systems suitable for resource-constrained environments, potentially improving early detection capabilities, and public health outcomes in developing regions.

  • Open Access

    Air quality prediction using boosting-based machine learning models for sustainable environment

    by Ahmad Fauzi, Maharina Maharina, Jamaludin Indra, Ayu Ratna Juwita, Agustia Hananto, Euis Nurlaelasari

    High levels of air pollution are extremely harmful to humans and the environment. They increase the risk of respiratory infections and lung cancer, especially among vulnerable populations. Therefore, developing effective pollution control measures is crucial for mitigating these negative impacts. We need to implement effective methods to predict and manage air quality for the sake of public health and a healthier environment. In recent years, machine learning (ML) methods have been increasingly utilized in air quality prediction due to their ability to analyze datasets and identify complex patterns. However, the reliability and accuracy of air quality prediction models remain a challenge. This study proposes a boosting-based ML model for predicting air quality. We implemented three stages in the proposed method. In the first stage, we conducted data preprocessing and analysis to eliminate noise, remove redundant data, and encode categorical features. In the second stage, we predicted air quality categories by leveraging 25 ML models, dividing them into three distinct categories. The results show that the extreme gradient boosting (XGBoost), light gradient boosting machine (LGBM), and adaptive boosting (AdaBoost) models outperform the others in air quality prediction, achieving an accuracy of 99%. Finally, we compared these three models using 10-fold cross validation to ensure they generalize well in last stage.

  • Open Access

    HyEWCos: A Comparative Study of Hybrid Embedding and Weighting Techniques for Text Similarity in Short Subjective Educational Text

    by Hendry Hendry, Tukino Tukino, Eko Sediyono, Ahmad Fauzi, Baenil Huda

    This study is intended to evaluate and contrast the performance of varying combinations of embedding algorithms and weighting systems in measuring perception-based text similarity using the Cosine Similarity approach. Within a structured experiment design, a hybrid model referred to as HyEWCos (Hybrid Embedding and Weighting for Cosine Similarity) was built incorporating conventional embedding models (Word2Vec, FastText), transformer-based models (BERT, GPT), and statistical and linguistic word weighting schemes (TFIDF, BM25, POS-weighting, and N-weighting). The test results indicate that Word2Vec merged with the CBOW architecture and TFIDF weighting always returned the most reliable performance, with lowest error values (RMSE and MAE of 0.9868) and the highest rating correlation with expert judgment (Pearson’s, 0.524; Spearman’s, 0.543). These results show that contextually conditioned distributional representation approaches perform better in maintaining the semantic subtlety of short and subjective texts than transformer models that are not fine-tuned. This work is unique in terms of its evaluation framework because it integrates embedding and weighting approaches that have hitherto been examined mostly in separation. The main contribution of the study is the development of an experimental framework that serves as a foundation for building more stable and accurate text-based assessment systems. The research also proves the need for making decisions on representation methods based on the data type and domain and opens a door for continuing research in adaptive hybrid models and how their potential can be achieved through combining the best of various approaches.


  • Open Access

    Analisis Sentimen Masyarakat Terhadap Pembatasan BBM Pertalite Menggunakan Random Forest dan K-Nearest Neighbor

    by Farhan Muhammad Fadillah, Yana Cahyana, Rahmat, Ahmad Fauzi

    This study aims to analyze public opinion regarding the policy of limiting the use of Pertalite fuel by examining user comments on the Instagram platform. To classify these opinions, classification approaches using K-Nearest Neighbor (KNN) and Random Forest algorithms were employed. Comments were categorized into three sentiment expressions: positive, negative, and neutral. The research stages included data collection (crawling), text cleaning and normalization, sentiment labeling, weighting using the TF-IDF technique, model development, and performance evaluation. A total of 2,081 comments were used, with 1,000 comments labeled by language experts as training data, and the remaining used for testing. Model evaluation was conducted using two data splitting ratios, 80:20 and 70:30, to assess classification stability and accuracy. The results indicate that the Random Forest algorithm consistently outperforms KNN, achieving the highest accuracy of 73% under the 80:20 scenario. The classification distribution suggests a dominance of negative sentiment in public opinion toward the policy. These findings reflect public dissatisfaction and serve as critical input for the government in reviewing the subsidized fuel distribution policy. This research also highlights the potential of social media as an alternative data source for real-time public perception analysis.


  • Open Access

    Road Damage Detection Using Yolov9-Based Imagery

    by Febrian Akbar Azhari, Tatang Rohana, Kiki Ahmad Baihaqi, Ahmad Fauzi

    Road damage is one of the leading factors contributing to traffic accidents. Rapid identification and repair of damaged roads are crucial in road infrastructure management. This study aims to develop an effective method for detecting road damage, utilizing the YOLOv9 algorithm as a key component, such as cracks and potholes, using the Convolutional Neural Network (CNN) approach. YOLOv9 was chosen due to its efficient architecture, which enables real-time object detection, and its proven effectiveness in various object detection tasks. An annotated dataset of road images was used during the model training and testing process. The results show that the YOLOv9 model can accurately detect road damage. The model achieved a precision of 0.85 and a recall of 0.992 for pothole detection, and a precision of 0.94 for crack detection. Evaluation using mAP50 yielded a score of 0.96, while mAP50-95 reached 0.77, indicating strong detection and classification capability. A consistent decline in loss functions during training also signifies effective learning by the model. These findings suggest that YOLOv9 has the potential to be implemented in automated road damage detection systems, which can accelerate maintenance processes and enhance road user safety.

  • Open Access

    PEMODELAN INSPEKSI PAINTING DEFECT PADA MOBIL MENGGUNAKAN CONVOLUTIONAL NEURAL NETWORK (CNN)

    by Muchamad Fachrul Ramadhan, Ahmad Fauzi, Deden Wahiddin, Tatang Rohana

    Quality control is an important process carried out at the last stage of the production process, this activity is carried out by checking a product. Painting defects on cars are a problem that must be considered in the car production process at car companies. The perfection of a product is important to increase the level of customer satisfaction. These checking activities are still carried out manually with human power, which can still cause defective products to be missed in a production process that occurs as a result of human error. The use of artificial intelligence can be used to detect image and video objects, used to overcome the problem of human error in carrying out checks. Convolutional Neural Networks (CNN) is an algorithm that can be used in product defect inspection, image recognition, and image classification. The study focuses on modeling the inspection and detection of painting defects in cars using CNN, emphasizing the importance of quality control in ensuring product quality. The CNN model is trained with image data of normal car paint and defective car paint, and evaluated using a confusion matrix for optimal parameters. The results show quite high accuracy in detecting car paint defects of 98% with the help of the ResNet50 transfer learning CNN architecture.

  • Open Access

    SEGMENTASI PELANGGAN MENGGUNAKAN K-MEANS CLUSTERING DI TOKO RETAIL

    by Syifa Latifah Achmad, Ahmad Fauzi, Rahmat Rahmat, Jamaludin Indra

    Advancements in information technology have transformed various aspects of human life, including the business world. Companies are required to use technology and data effectively to enhance their competitive advantage. One increasingly relevant strategy is Customer Relationship Management (CRM), where customer data is the main focus. Consumer data segmentation is an approach used to group customers based on certain characteristics. In this study, the K-Means Clustering algorithm is applied to consumer data segmentation to improve the marketing strategy of a store. The study begins with the collection of customer data from the Dan+Dan Telukjambe 2 store, followed by Exploratory Data Analysis (EDA) to understand the patterns and characteristics of the data. Preprocessing steps are carried out to ensure the data is ready for use, including removing irrelevant columns, handling missing values, and data transformation. Principal Component Analysis (PCA) is used to reduce data dimensions before applying K-Means Clustering. The Elbow Method and Silhouette Score are used to determine the optimal number of clusters. The study results indicate that the optimal number of clusters is six. Evaluation using the Silhouette Coefficient provides an average coefficient value of 0.66, indicating good clustering quality. Further analysis shows different distributions of age, purchasing power, occupation, and marital status in each cluster, providing deep insights into customer segments. The resulting clusters offer valuable information for developing more effective and targeted marketing strategies


  • Open Access

    Implementasi Hyperparameter Tuning Grid Search CV Pada Prediksi Produksi Padi Menggunakan Algoritma Linear Regresi

    by Fathir Jamiluddin, Sutan Faisal, Santi Arum Puspita Lestari, Ahmad Fauzi

    Rice is one of the main crops in Indonesia that produces the largest staple food, namely rice commodities. Rice is a staple food consumed by almost 98% of Indonesian people. This study aims to compare the Linear Regression Algorithm and Decision Tree in an effort to find the most appropriate algorithm for predicting rice production data. Linear Regression is still a useful model, especially if the data has a non- linear relationship that cannot be captured by Linear Regression. So it can be concluded that the Linear Regression Algorithm with optimization of the tuning grid search cv hyperparameter is able to predict rice production better than the Decision Tree Algorithm with an R2-score value of 86.895666, MAE 261049.168107, and MSE 160199780301.226318.


  • Open Access

    KLASIFIKASI PECAHAN UANG KERTAS RUPIAH MENGGUNAKAN TRANSFER LEARNING DENGAN MODEL MOBILENETV2

    by Rissa Ilmia Agustin, Jamaludin Indra, Sutan Faisal, Ahmad Fauzi, Rija Nur Hijriyya

    Memanfaatkan mesin sebagai perantara dalam proses pembelian dan penjualan adalah bagaimana teknologi otomasi diterapkan. Mesin berfungsi sebagai penjual dan memiliki kemampuan seperti otak, seperti kecepatan dan keakuratan dalam membaca dan mengidentifikasi nominal uang. Dengan menggunakan teknologi otomatis ini, transaksi jual beli menjadi lebih nyaman. Metode Convolutional Neural Network (CNN) merupakan salah satu komponen dari teknologi Deep Transfer Learning digunakan dalam penelitian ini untuk mengenali uang kertas rupiah. Selain itu, penelitian ini memilih arsitektur model MobileNetV2 yang sesuai dan memodifikasi laju pembelajaran keduanya berdampak pada kinerja model klasifikasi. Untuk menjamin bahwa model memiliki kesempatan yang memadai untuk belajar dari data pelatihan, jumlah epoch yang ideal juga diperhitungkan. Selain itu, hal ini dapat berdampak pada pencapaian kinerja tinggi dengan waktu komputasi yang efisien, pemanfaatan ukuran batch yang optimal juga diselidiki. Evaluasi kinerja model selama pelatihan memberikan hasil sebagai berikut : f1-score 98% recall 98%, presisi 98%, dan akurasi pada set pengujian 97.86%.


  • Open Access

    Perbandingan Metode K-Means dan K-Medoids Untuk Clustering Jenis Kriminalitas

    by Nurul Azizah, Ahmad Fauzi, Tatang Rohana, Sutan Faisal

    Crime in Indonesia includes acts that violate the law, social norms and religion which cause economic and psychological losses as well as social tensions in society. Crimes such as theft, violence, fraud and drugs are often triggered by factors such as poverty and environmental conditions that support criminal behavior. This research needs to be carried out to overcome the complex and far-reaching crime problem in Indonesia, especially in Karawang Regency. With crimes such as theft, violence, fraud and drugs on the rise, often fueled by factors such as poverty and environmental conditions, a more effective approach is needed to understand and address these problems. This research uses data mining techniques, especially cluster analysis, to group types of crime. The aim is to identify existing crime patterns and understand the factors that influence their spread. Thus, the results of this research can help the authorities in developing more targeted crime prevention and handling strategies, so as to minimize the negative impact of crime in the area. Apart from that, this research also contributes to increasing knowledge regarding the most effective methods for analyzing crime data, which can be applied in other areas with similar problems. The results of the research show that the K-Means algorithm is more effective than K-Medoids in handling data variability, with a Silhouette Coefficient value of 0.482 and a Davies Bouldin Index of 0.915. It is hoped that the implementation of this algorithm will make it easier to identify and handle crimes in the area.