Bulletin of Information Technology (BIT)

Optimalisasi Random Forest dengan Penyelarasan Temporal untuk Identifikasi Faktor Determinan Stunting Nusa Tenggara Barat

2026-06-23T01:55:33+00:00

Stunting remains a major public health issue in West Nusa Tenggara, with key challenges including low data integrity caused by reporting discontinuities, spatial heterogeneity, and statistical noise that obscure accurate determinant identification. These issues are particularly evident during the 2023–2024 reporting transition, where inconsistencies distort the relationship between community health worker performance and stunting prevalence. This study applies a Spatio-temporal approach incorporating temporal alignment, normalization, and Random Forest regression. The analysis includes Moderated Regression Analysis using Ordinary Least Squares, R², and Global Moran's I. The objective is to improve accuracy in identifying determinants of stunting reduction and evaluate community health worker effectiveness using integrated data. Results show consistent decline in stunting prevalence during 2018–2024, with significant reductions in Central of Lombok (-42.2%), West Lombok (-39.4%), and Bima City (-47.7%), while increases occurred in Bima Regency (+8.5%) and Mataram City (+33.6%). Model performance improves significantly after data alignment, with R² exceeding 0.70. Feature importance identifies cadre workload ratio and child weighing participation as dominant predictors (p = 0.073), indicating a discovery effect. Spatial analysis yields Global Moran's I of -0.0615, suggesting a dispersed pattern. Overall, integrating temporal data alignment with machine learning and spatial analysis enhances determinant identification and supports data-driven policy for stunting reduction in West Nusa Tenggara.

Pengelompokan Data Penjualan Produk Cetakan Pada Algoritma K-Means Dengan Bantuan Tool Orange

2026-06-23T01:55:33+00:00

The main problem faced is the large amount of unstructured sales data, making it difficult to perform manual analysis. With the application of the K-Means algorithm, sales data can be grouped into clusters representing products with high and low sales. The research process begins with the stages of problem identification, data collection, preprocessing, application of the K-Means algorithm, evaluation of clustering results, and then analysis and interpretation. Iteration results show that cluster C1 consists of a number of high-selling sales data, while cluster C2 encompasses the majority of low-selling sales data. Evaluation using the Davies-Bouldin Index (DBI) yields a value of 0.2818, indicating fairly good cluster quality, while the Silhouette Plot provides values of 0.082 for C1 and 0.276 for C2, indicating that cluster C2 is more stable compared to C1. Scatter Plot visualization shows the data distribution forming a slanted pattern from C1 to C2. The result of this research is that by using the K-Means algorithm, it can effectively cluster sales data of printed products, so it can be used as a basis for business decision-making related to marketing strategies, stock control, and product performance evaluation.

Analisis Penentuan Pengolahan Kopi Arabika Terbaik Dengan Metode ELECTRE

2026-06-23T01:55:33+00:00

Arabica coffee is a premium commodity with high economic potential, while also offering distinctive taste and aroma that make it valuable in both national and international markets. In processing practice, choosing the right method becomes its own challenge, considering the various alternatives such as wet, dry, and semi-wet methods that must be evaluated based on quality, cost, and environmental impact. The main issue lies in how to determine the optimal processing method that can produce coffee beans of superior quality. This study uses the ELECTRE method, one of the multi-criteria decision-making approaches, to evaluate various Arabica coffee bean processing alternatives. The analysis process includes the preparation of a decision matrix, normalization, weighting of criteria (aroma, flavor, aftertaste, acidity, body, and balance), calculation of concordance and discordance matrices, as well as dominance analysis to obtain priority. The calculation results show that alternative 6 (A6) for Pulped Natural/Honey and A10 for Wet Hulling are the best choices for processing Arabica coffee beans. These findings provide practical solutions for farmers and coffee industry players in improving quality as well as the competitiveness of Indonesian Arabica coffee.

Clustering Status Gizi Menggunakan Algoritma K-Means Dengan Pendekatan CRISP-DM

2026-06-23T01:56:29+00:00

The nutritional status of toddlers is a key indicator for assessing public health levels. Nutritional problems such as undernutrition remain common and require effective analysis to identify data patterns quickly and accurately. This study aims to cluster the nutritional status of toddlers in the Bukuan Community Health Center (Puskesmas) Posyandu area using the K-Means algorithm with the CRISP-DM approach. The main challenge in this study is that the processing of nutritional status data is still done manually, making it less effective in quickly and accurately identifying patterns and risk groups. The dataset consists of 2,145 records of toddlers from 12 Posyandu, with primary attributes including weight, height, age, and gender. The research process was conducted through the CRISP-DM stages, which include business understanding, data understanding, data preparation, modeling, and evaluation, without deployment implementation since the study focused on data analysis. The clustering process was performed using the K-Means algorithm, with the optimal number of clusters determined via the Elbow method, resulting in three clusters. Model evaluation using the Silhouette Score yielded a value of 0.629, indicating that the clustering quality falls into the “good” category. The results show that data on toddlers can be grouped into three nutritional status categories: under-nutrition, adequate nutrition (normal), and over-nutrition, based on centroid values. The data distribution indicates that the adequate nutrition category dominates, though there remains a significant number of cases in the under-nutrition category. Thus, the application of the K-Means algorithm provides more structured and accurate information for identifying the nutritional status of toddlers and can serve as a basis for data-driven decision-making in public health programs.

Pemanfaatan Algoritma K-Medoids Clustering dalam Menentukan Pendapatan Bersih Komoditas Pertanian

2026-06-23T03:06:43+00:00

Agricultural products are one of the sectors that have a major role in the Indonesian economy. Currently, Indonesia is the largest producer in the world that produces Palm Oil, Cloves, Cinnamon, Nutmeg, and many others. Abundant agricultural products can be applied to research using Data Mining techniques. Data Mining is a technique that applies statistical analysis and artificial intelligence in extracting useful information from a database. In this study the author will use the K-Medoids method, K-Medoids is one of the Data Mining techniques. Analysis of K-Medoids results uses the silhouette coefficient which is used to measure the distance between clusters. The objective value using K-Medoids cluster analysis on the dataset used is 5.742047 and 5.093438. After conducting cluster analysis with the silhouette coefficient, the best results obtained are 2 clusters from 12 data and 12 attributes.

Pengembangan Model Deteksi Autism Spectrum Disorder (ASD) Dengan Algoritma Facenet Vggface Dan Insightface

2026-06-23T04:21:04+00:00

Autism Spectrum Disorder (ASD) is a neurodevelopmental condition that affects communication, social interaction, and behavior. Conventional ASD diagnosis relies on clinical observation, which is time-consuming and subjective. Therefore, an automated approach using artificial intelligence is required to support early detection. This study proposes an ASD detection model based on facial image analysis using deep learning approaches, namely FaceNet, VGGFace2, and InsightFace as facial feature extraction methods. The dataset consists of 3,620 facial images categorized into ASD and non-ASD classes. The research process includes preprocessing, feature extraction, model training, and evaluation using accuracy, precision, recall, and F1-score metrics. The results indicate that all models achieved good classification performance, with FaceNet achieving the highest accuracy of 98%, followed by InsightFace with 96%, and VGGFace2 with 95%. These findings demonstrate that face embedding-based models provide superior feature extraction capabilities for ASD detection.

Analisis Sentimen Masyarakat Terhadap Program Gratis Pol Di TikTok Menggunakan Algoritma Naive bayes

2026-06-23T04:39:05+00:00

The Gratis Pol Program is an educational initiative in East Kalimantan that has garnered public attention and sparked a wide range of reactions on social media, particularly TikTok. The characteristic use of informal language, abbreviations, and colloquial expressions in TikTok comments poses a challenge for sentiment analysis, necessitating a method capable of systematically classifying public opinion. This study aims to analyze public sentiment toward the Gratis Pol Program based on TikTok user comments using the Naive Bayes algorithm. The research was conducted through the stages of text preprocessing, sentiment labeling using a lexicon-based approach, feature representation using TF-IDF, and the classification process using the Naive Bayes algorithm. The research data was obtained from 13 selected TikTok videos with a total of 1,528 comments, divided into 80% training data and 20% test data. The results show that positive sentiment dominates with 722 comments, followed by 496 neutral comments and 310 negative comments. The classification model achieved an accuracy of 56%, with a macro average F1-score of 0.44 and a weighted average F1-score of 0.48. This study contributes to understanding public perception of the Gratis Pol Program and demonstrates the application of the Naive Bayes algorithm in analyzing the sentiment of social media comments that possess certain characteristics.

Implementasi Klasifikasi Teks Menggunakan Algoritma Naïve Bayes pada Sistem Pengarsipan Surat Masuk

2026-06-23T05:10:46+00:00

Incoming mail management is an important part of higher education administration because it is related to document storage, classification, and retrieval in a fast and accurate manner. However, the incoming mail archiving process at the General Administration and Finance Bureau (BAUK) of STMIK Widya Cipta Dharma is still carried out manually, making document grouping inefficient, slowing down archive retrieval, and potentially causing inconsistencies in determining mail categories. This study aims to implement the Naïve Bayes algorithm in a web-based incoming mail archiving system to support automatic mail classification. The system was developed using Laravel and Livewire as the main application, and Flask as the classification service. The dataset consisted of 79 incoming mail documents divided into four categories: requests, invitations, notifications, and reports. The preprocessing stage included case folding, text cleaning, tokenizing, stopword removal, and stemming. The results show that the system is able to automatically classify incoming mail and present detailed classification processes through training reports and classification results. Based on testing on 16 incoming mail documents, the model achieved an accuracy of 75.00%, an average precision of 63.89%, and an average recall of 72.22%. These results indicate that the Naïve Bayes algorithm is sufficiently effective in supporting a more structured and efficient incoming mail archiving process.

Optimasi Penentuan Sales Ececutive Terbaik Menggunakan Metode MOORA Pada Dealer Mitsubishi Tenggarong

2026-06-23T07:13:14+00:00

Penentuan sales executive terbaik merupakan salah satu upaya penting dalam meningkatkan kinerja dan daya saing perusahaan, khususnya pada dealer otomotif, karena berpengaruh langsung terhadap pencapaian target penjualan dan kualitas pelayanan pelanggan. Permasalahan yang sering terjadi adalah proses penilaian yang masih bersifat subjektif dan belum menggunakan metode yang terstruktur, sehingga hasilnya kurang objektif dan konsisten. Penelitian ini bertujuan untuk membangun sistem pendukung keputusan dalam menentukan sales executive terbaik dengan menggunakan metode Multi-Objective Optimization on the Basis of Ratio Analysis (MOORA) sebagai solusi berbasis pengambilan keputusan multikriteria. Data yang digunakan merupakan data internal dealer Mitsubishi yang mencakup beberapa alternatif sales executive dengan kriteria penilaian berupa pencapaian target penjualan, kedisiplinan, kemampuan komunikasi, dan kualitas pelayanan. Proses pengolahan data dilakukan melalui tahap pemilihan data, pra-pemrosesan data, penentuan kriteria pembobotan, implementasi metode, hingga perangkingan. Hasil penelitian menunjukkan bahwa metode MOORA mampu menghasilkan nilai preferensi dan urutan peringkat sales executive secara objektif dan sistematis. Alternatif dengan nilai tertinggi ditetapkan sebagai sales executive terbaik. Dengan demikian, sistem yang dibangun dapat membantu pihak manajemen dalam mengambil keputusan yang lebih efektif, efisien, dan transparan serta mendukung evaluasi kinerja berbasis data yang lebih optimal.

Implementasi Robot Mobile Dual Mode Berbasis Arduino Uno sebagai Media Pembelajaran Ekstrakulikuler Robotik

2026-06-23T07:45:09+00:00

The advancement of robotics technology offers significant opportunities to enhance practice-based learning in vocational education. However, the implementation of robotics learning at the vocational high school (SMK) level still faces limitations in interactive learning media and low student engagement. This study aims to develop and evaluate a dual-mode mobile robot based on Arduino Uno as a learning medium for robotics extracurricular activities at SMK Negeri 11 Kabupaten Tangerang. The research method employs a Research and Development (R&D) approach with a prototyping model, including problem identification, system design, implementation, testing, and evaluation stages. The developed system integrates two operational modes: an automatic mode (obstacle avoider) and a manual control mode using Bluetooth communication. The results indicate that the system operates stably and responsively, with an average response time of less than one second and acceptable sensor accuracy. Learning evaluation through questionnaires shows an improvement in student interest and understanding, with scores above 78%. This study contributes to the development of interactive and practical robotics-based learning media, which enhances student engagement in the learning process.

Optimasi Random Forest Melalui Feature Engineering dan SMOTE untuk Klasifikasi Kesehatan Mental

2026-06-23T07:45:30+00:00

Student mental health is a crucial factor affecting academic performance, productivity, and overall quality of life in university environments. The high prevalence of psychological disorders today demands an accurate early detection system to provide timely and efficient intervention. This study aims to develop a student mental health classification model by integrating feature engineering techniques and the Synthetic Minority Oversampling Technique (SMOTE) with the Random Forest algorithm. The feature engineering stage is conducted through the creation of a composite feature, Mental_Score, to represent students' psychological conditions more holistically and deeply. In addition, SMOTE is applied to address the data imbalance issue, making the model more sensitive in detecting the at-risk student group as the minority class. Experimental results show that the proposed model achieves an accuracy of 97%. The application of SMOTE proved effective in increasing the minority class recall to 60% and raising the F1-score from 0.57 to 0.75, significantly strengthening the detection capability for the at-risk group. Although the McNemar test yields a p-value of 1.000 due to a ceiling effect since both models are already optimal, the proposed model still offers a practical advantage in maintaining detection sensitivity. Feature importance analysis confirms that Mental_Score is the most influential attribute with a contribution value of 0.3280. This study contributes to providing a more accurate machine learning-based framework for the early detection of student mental health.

Komparasi Metode EUCS dan TAM Dalam Analisis dan Implementasi Sistem E-commerce

2026-06-23T08:22:50+00:00

Tujuan penelitian ini adalah untuk menganalisis dan membandingkan tingkat penerimaan dan kepuasan pengguna sistem e-commerce SMK Bina Patriot. Kurangnya evaluasi komprehensif terhadap sistem yang diterapkan, terutama terkait dengan kebahagiaan pengguna dan penerimaan teknologi, merupakan perhatian utama yang disoroti oleh penelitian ini. User Acceptance Model (TAM) dan End User Computing Satisfaction (EUCS) adalah dua metodologi yang digunakan. Dalam mengukur kepuasan pengguna, EUCS menggunakan lima kriteria: konten, akurasi, format, kemudahan penggunaan, dan ketepatan waktu. Di sisi lain, TAM menggunakan persepsi pengguna tentang utilitas yang dirasakan untuk menentukan penerimaan teknologi. Data dikumpulkan melalui survei yang dikirimkan kepada siswa melalui sistem. Temuan penelitian ini mendukung reliabilitas EUCS (0,854) dan TAM (0,865), menunjukkan bahwa instrumen ini dapat dipercaya. Tingkat kepuasan yang sangat tinggi ditunjukkan oleh hasil analisis deskriptif, yang menunjukkan skor EUCS rata-rata 4,16 dan skor TAM 4,22. Dalam studi terhadap kedua variabel tersebut ditemukan hubungan yang sangat kuat dan signifikan secara statistik sebesar 0,805. Menurut studi regresi, persepsi pengguna terhadap kegunaan produk dipengaruhi secara positif oleh kemudahan penggunaannya. Studi ini menggunakan pendekatan perbandingan terhadap dua model, yang berkontribusi pada evaluasi sistem e-commerce berbasis pendidikan.

Optimasi Model Yolov8n Menggunakan Augmentasi Data Untuk Peningkatan Akurasi Sistem Dress-Code Surveillance

2026-06-23T09:13:21+00:00

Manual surveillance of student dress-code compliance on campus is often inefficient, subjective, and constrained by the physical fatigue of security personnel. This study aims to automate the surveillance system by optimizing the Nano variant of the YOLOv8 (YOLOv8n) Deep Learning model based on Computer Vision. The main challenge in real-time object detection is limited datasets and visual diversity, which increases the risk of overfitting. The solution applied to address this issue is the implementation of comprehensive dynamic data augmentation strategies, including Hue, Saturation, Value (HSV) manipulation, Horizontal Flipping, and Mosaic Augmentation. Utilizing the CRISP-DM methodology, this technique expanded the dataset from 5,000 initial images to 9,742 training images. The empirical test results show that the optimized YOLOv8n model significantly improved accuracy by 43,5% compared to the baseline model. The best-performing model achieved a Mean Average Precision (mAP@0.5) of 95.3%, with a Precision of 93.1%, Recall of 91.1%, and an F1-score of 0.92. These metrics demonstrate the reliability of the system in reducing false positives while operating in crowded real-world environments. This automated surveillance system is highly feasible for direct integration into campus CCTV infrastructure using edge computing to objectively support institutional discipline.

Optimalisasi Pengenalan Wajah Pada Kondisi Low Light Menggunakan YOLOv5 Face Dan CLAHE

2026-06-23T09:26:44+00:00

Degradation of illumination intensity in low light environments is a major issue that reduces the accuracy of deep learning based face detection and recognition systems. This study aims to optimize the performance of the biometric processing pipeline under extreme low light conditions without retraining the model. The novelty of this research lies in the design of a retrainless integration between the hybrid preprocessing of CLAHE and Bilateral Filter on the luminance channel of the LAB color space with the integrated processing pipeline of YOLOv5-Face and FaceNet. Mass testing was conducted using pair-based evaluation on 3,000 face pairs from the VGGFace2 benchmark dataset, simulated in a controlled manner using a gamma exponent value (y = 0,5). Experimental results show that the preprocessing stage successfully restored the YOLOv5-Face Detection Rate from 88.7% to 89.8%. Meanwhile, in the identity verification stage, the FaceNet model recorded an increase in class separability, achieving the highest Area Under the Curve (AUC-ROC) value of 0.927 (classified as excellent), a global accuracy of 89.3%, and the ability to maintain the stability of the Cosine Distance Gap at an index of 0.6005. This characteristic proves the robustness of the feature vector geometry in separating boundaries between identities without overlapping. The implementation of the system into a Streamlit web application confirms that this traditional contrast restoration method remains relevant, efficient, and reliable for securing biometric verification under low-light conditions.

Analisis Business Intelligence Keluhan Merchant GoFood: Identifikasi Tema dan Evolusi Keluhan Menggunakan BERTopic

2026-06-23T09:41:20+00:00

Platform Online Food Delivery (OFD) telah menjadi bagian penting dari ekonomi digital dengan memungkinkan merchant memperluas jangkauan pasar dan meningkatkan efisiensi operasional bisnis. Namun, meningkatnya ketergantungan pada ekosistem yang dikelola platform juga menimbulkan berbagai tantangan operasional dan bisnis bagi para merchant. Penelitian ini bertujuan untuk mengidentifikasi topik utama keluhan merchant GoFood, menganalisis evolusi temporal topik tersebut, serta menghasilkan rekomendasi strategis dari perspektif Business Intelligence. Data penelitian terdiri atas 7.013 ulasan negatif yang dikumpulkan dari aplikasi GoFood Merchant di Google Play Store selama periode 1 Juni 2023 hingga 30 Mei 2026. BERTopic digunakan untuk mengidentifikasi topik-topik utama yang terkandung dalam ulasan merchant, sedangkan analisis temporal dilakukan untuk mengamati perubahan frekuensi kemunculan topik dari waktu ke waktu. Hasil evaluasi model menunjukkan nilai Topic Coherence sebesar 0,590, Topic Diversity sebesar 0,611, dan Topic Quality sebesar 0,360. Hasil penelitian mengidentifikasi 12 topik utama keluhan merchant, dengan biaya promosi, permasalahan pengemudi, permasalahan keuangan, dan ketidakpuasan terhadap kebijakan platform sebagai topik yang paling dominan. Analisis temporal menunjukkan bahwa topik-topik tersebut tetap menjadi isu utama selama periode observasi. Dari perspektif Business Intelligence, temuan penelitian mengungkap empat dimensi kerentanan utama dalam ekosistem, yaitu tekanan monetisasi, ketergantungan operasional, kerentanan finansial, dan ketidakpuasan terhadap tata kelola platform. Temuan ini dapat mendukung pengelola platform dalam meningkatkan kualitas layanan, kepuasan merchant, serta keberlanjutan ekosistem platform.

Klasifikasi Kematangan Buah Pinang (Areca catechu L.) Menggunakan Hybrid Deep Feature Fusion dan XGBoost

2026-06-23T10:06:56+00:00

Areca nut (Areca catechu L.) maturity is one of the factors affecting harvest quality. Visual maturity identification still has limitations because it can be influenced by observer subjectivity and environmental conditions. This study aims to classify areca nut maturity levels using a Hybrid Deep Feature Fusion approach by combining ResNet50 and EfficientNetB0 as feature extractors with XGBoost as the classification algorithm. The dataset used in this study was a primary dataset consisting of 1,200 areca nut images categorized into three maturity classes: unripe, semi-ripe, and ripe. The research stages included image preprocessing, feature extraction using CNN models, feature combination through feature concatenation, classification using XGBoost, and performance evaluation using accuracy, precision, recall, F1-score, confusion matrix, and 5-Fold Cross Validation. The experimental results showed that ResNet50 + XGBoost and Hybrid Deep Feature Fusion + XGBoost achieved accuracy, precision, recall, and F1-score values of 100%, while EfficientNetB0 + XGBoost achieved an accuracy of 99.16%. These results indicate that CNN-based features are able to represent the visual characteristics of areca nut images in the dataset used. The Hybrid Deep Feature Fusion approach provides an analysis of feature combination from two different CNN architectures, although increasing the feature dimensions does not always improve evaluation performance when a single feature extractor is already capable of representing dataset characteristics effectively. Future research can be conducted by increasing dataset variations to evaluate the generalization capability of the method under more diverse environmental conditions.

Perbandingan SVM dan IndoBERT untuk Analisis Sentimen Layanan Akademik Mahasiswa

2026-06-27T07:58:12+00:00

Digital transformation in higher education has generated an increasing volume of textual data, including student comments, academic service evaluations, and feedback on academic information systems. These data contain valuable information for supporting decision-making; however, their unstructured and contextual nature makes manual analysis inefficient. This study aims to compare the performance of a TF-IDF-based Support Vector Machine (SVM) model and a Transformer-based IndoBERT model for sentiment analysis of academic services from student feedback. The dataset consists of 1,700 text entries, combining template-based synthetic data and real-world data collected from social media, which were classified into positive, negative, and neutral sentiment categories. The research process involved exploratory data analysis, text preprocessing, feature extraction, model development, and evaluation using accuracy, precision, recall, and F1-score metrics. The experimental results showed that both models achieved very high performance on the dataset, with an accuracy of 100% on the test set. These findings indicate that both traditional machine learning and Transformer-based approaches are capable of identifying sentiment patterns within the dataset. Nevertheless, the results should be interpreted cautiously, as the relatively homogeneous nature of the dataset and the inclusion of synthetic data may affect the models’ generalizability. The main contribution of this study lies in the comparative evaluation of SVM and IndoBERT within the context of higher education academic services, as well as the development of a sentiment analysis framework that can support data-driven service quality monitoring. Future studies should employ larger, more diverse datasets derived entirely from real-world sources to further validate the findings.

Implementasi Metode SAW Dan MAUT Dalam Sistem Pendukung Keputusan Menentukan Verietas Nanas Terbaik

2026-06-27T08:10:11+00:00

This research aims to assist farmers and agribusiness practitioners in determining the best pineapple variety in a more objective and systematic manner. Ultimately, this will positively impact the productivity and quality of pineapple yields, providing greater economic benefits to farmers and agribusiness practitioners. To address this issue, a Decision Support System (DSS) was employed using the SAW and MAUT methods. The SAW method is a simple and easily implemented MCDM method. It works by assigning weights to each criterion and then calculating the total score for each alternative based on these weights. On the other hand, the MAUT method is a more comprehensive approach to multi-criteria decision-making. This method is based on utility theory, which considers the decision-maker's preferences regarding various attributes or criteria. By using the SAW and MAUT methods, we can determine the best pineapple variety selection based on the available data. From the previously collected data, this study uses 5 criteria: Size (30%), Taste (25%), Skin Color (20%), Water Content (15%), and Texture (10%). The implementation of the SAW and MAUT methods revealed that the best pineapple variety is A5 (MD2) with a score of 0.9125 using the SAW method and a score of 2.2062 using the MAUT method. The last-ranked alternative, A10, shared the same ranking.

Evaluasi YOLOv8 Nano Untuk Deteksi Logistik Pendaki Pada Clutter Ekstrem

2026-07-03T01:56:48+00:00

Pemeriksaan logistik pendaki gunung secara manual saat ini dinilai tidak efisien dan sangat rentan terhadap kesalahan manusia (human error) akibat tingginya volume serta variasi barang bawaan yang sering menumpuk (clutter). Meskipun algoritma YOLOv8 sangat populer untuk deteksi objek, kinerjanya pada skenario kepadatan visual ekstrem belum teruji secara komprehensif. Oleh karena itu, penelitian ini bertujuan untuk mengevaluasi ketangguhan model YOLOv8 Nano dalam mengidentifikasi lima kelas logistik pendaki pada lima tingkat kepadatan, mulai dari objek tunggal hingga tumpukan ekstrem. Penelitian ini mengadopsi metodologi standar CRISP-ML(Q) dengan memanfaatkan 13.792 sampel data kustom. Fase prapemrosesan menerapkan metode Stretch to guna mereduksi artefak visual pada area tepi citra. Hasil eksperimen mendemonstrasikan performa yang sangat presisi, ditandai dengan nilai Precision sebesar 0,971, Recall 0,954, dan mean Average Precision (mAP@50) mencapai 97,8%. Arsitektur ini terbukti sanggup mendobrak limitasi penelitian terdahulu dengan keberhasilan mempertahankan stabilitas mAP@50 di angka 96,22% pada pengujian kepadatan ekstrem (lebih dari 18 objek). Implementasi sistem berbasis aplikasi web lintas perangkat juga mencatatkan waktu inferensi real-time yang responsif, yakni 61,48 milidetik pada peramban laptop dan 72,62 milidetik pada telepon seluler. Kesimpulannya, algoritma YOLOv8n terbukti sangat reliabel untuk mengotomatisasi pelaporan logistik lapangan. Namun, limitasi masih ditemukan berupa degradasi akurasi pada objek mikro akibat fenomena kemiripan fitur antar-kelas dan distorsi pantulan cahaya. Studi mendatang direkomendasikan untuk mengintegrasikan teknik Slicing Aided Hyper Inference (SAHI) guna memitigasi kegagalan tersebut.