Jurnal Ilmu Komputer dan Agri-Informatika
Not a member yet
160 research outputs found
Sort by
Application of K-Nearest Neighbor Method and Support Vector Machine for Noni Fruit Ripeness Classification
Full text linkBuah mengkudu (Morinda citrifolia) merupakan salah satu komoditas ekspor buah-buahan di Indonesia yang selalu tersedia di setiap musim dan dikenal memiliki berbagai manfaat kesehatan. Buah mengkudu berasal dari wilayah Asia Tenggara, termasuk Indonesia, dan sering digunakan dalam pengobatan tradisional. Pada umumnya masyarakat menentukan kematangan buah mengkudu secara manual, yaitu dengan menggunakan penampakan visual. Hal ini menyebabkan adanya perbedaan persepsi dalam menentukan tingkat kematangan buah mengkudu. Oleh karena itu, penelitian ini bertujuan membangun model machine learning untuk klasifikasi tingkat kematangan buah mengkudu. Metode klasifikasi yang digunakan adalah K-Nearest Neighbor (KNN) dan Support Vector Machine (SVM) dengan menggunakan ekstraksi fitur warna Hue Saturation Intensity (HSI) dan ekstraksi fitur tekstur Local Binary Pattern (LBP). Pengklasifikasian yang dilakukan pada buah mengkudu dengan algoritma KNN menghasilkan model klasifikasi yang lebih baik daripada menggunakan algoritma SVM. Akurasi terbaik yang dihasilkan oleh KNN sebesar 88.62% pada k=11, sedangkan akurasi terbaik SVM dengan kernel polynomial sebesar 87.80%, menggunakan parameter C=0.1 Gamma=1, Degree=5, dan coef0=1.0. Hasil ini didapatkan dari data latih dan data uji dengan perbandingan 80:20.Noni fruit (Morinda citrifolia) is one of Indonesia’s export commodities. It is available year-round and is well known for its numerous health benefits. Native to Southeast Asia, including Indonesia, noni fruit is widely used in traditional medicine. Typically, the ripeness of noni fruit is determined manually based on visual inspection, which can lead to subjective judgments and inconsistent results. Therefore, this study aims to develop a machine-learning model to classify the ripeness levels of noni fruit. The classification methods employed are K-Nearest Neighbor (KNN) and Support Vector Machine (SVM), utilizing Hue-Saturation-Intensity (HSI) color features and Local Binary Pattern (LBP) texture features. Experimental results show that the KNN algorithm outperforms the SVM algorithm in terms of classification accuracy. The highest accuracy achieved using KNN was 88.62% at k = 11, whereas the best accuracy obtained with SVM using a polynomial kernel was 87.80%, with parameters set to C = 0.1, Gamma = 1, Degree = 5, and coef0 = 1.0. These results were achieved using an 80:20 split ratio for training and testing data
Feature Extraction and Visual Classification Methods for Identifying the Quality of Local Food Based on Digital Images
Full text linkMutu pangan lokal memiliki peran penting dalam menjaga ketahanan pangan dan meningkatkan daya saing produk di pasar. Namun, penentuan mutu masih mengandalkan inspeksi visual manual yang bersifat subjektif dan tidak konsisten, sehingga berpotensi menimbulkan kerugian pascapanen serta menurunkan kepercayaan konsumen. Penelitian ini bertujuan untuk mengembangkan sistem klasifikasi mutu pangan berbasis citra digital dengan memanfaatkan arsitektur deep learning EfficientNetV2B0 melalui pendekatan transfer learning. Dataset yang digunakan mencakup delapan kelas dari empat komoditas utama, yaitu apel, pisang, tomat, dan pare dalam kondisi segar dan tidak segar. Seluruh citra diproses melalui tahap prapengolahan berupa pengubahan ukuran dan normalisasi, serta dilakukan augmentasi data untuk meningkatkan variasi dan mencegah overfitting. Model dilatih dengan konfigurasi ringan dan dievaluasi menggunakan akurasi, presisi, recall, F1-score, serta confusion matrix. Hasil penelitian menunjukkan bahwa model mampu mencapai akurasi keseluruhan sebesar 99.8 persen dengan nilai presisi, recall, dan F1-score mendekati sempurna pada hampir semua kelas. Temuan ini membuktikan bahwa penerapan transfer learning dan augmentasi data efektif dalam membedakan pangan segar dan tidak segar berdasarkan citra digital. Secara praktis, sistem ini berpotensi mendukung petani dan pelaku usaha kecil menengah dalam melakukan inspeksi mutu yang lebih objektif, efisien, dan konsisten.Local food quality is essential for ensuring food security and enhancing product competitiveness in the market. Despite its importance, quality assessment predominantly depends on manual visual inspection, which is inherently subjective and inconsistent. This reliance contributes to post-harvest losses and diminished consumer trust. This study aims to develop a food quality classification system based on digital images by employing the EfficientNetV2B0 deep learning architecture through a transfer learning approach. The dataset consists of eight classes from four major commodities, namely apples, bananas, tomatoes, and bitter gourds, each in fresh and non-fresh conditions. All images were pre-processed through resizing and normalization, followed by data augmentation to increase variability and mitigate overfitting. The model was trained using a lightweight configuration and evaluated with accuracy, precision, recall, F1-score, and confusion matrix metrics. The results show that the proposed model achieved an overall accuracy of 99.8 percent, with precision, recall, and F1-score values approaching perfection across most classes. These findings demonstrate that the combination of transfer learning and data augmentation is effective in distinguishing between fresh and non-fresh food products based on visual features. In practical terms, this system has potential to support farmers and small-medium enterprises in conducting more objective, efficient, and consistent quality inspection
Topic Modelling on Beauty Product Reviews Using Latent Dirichlet Allocation
Full text linkIn contemporary society, beauty products have become essential, particularly for women. With their growing popularity, online review platforms now provide extensive information on product trends, customer satisfaction, and performance. However, the sheer volume of available reviews presents challenges in drawing meaningful conclusions. To address this, topic modeling techniques such as Latent Dirichlet Allocation (LDA) have been widely employed in text mining and information retrieval. LDA is a probabilistic model capable of uncovering latent structures within textual data and identifying similarities across documents. Recent studies suggest that topic modeling of product reviews in the cosmetics industry can yield valuable insights into consumer perceptions and product attributes. This study aims to identify thematic patterns in customer reviews of ten facial cleanser brands sourced from the Female Daily website. The research methodology consists of five main stages: data collection, preprocessing, topic modeling using LDA, visualization, and topic interpretation. The results reveal that Topic 2, which highlights preferred product advantages, is the most frequently discussed, accounting for 48.5% of the total reviews. Topic 1, which focuses on the effects of products on acne-prone skin, constitutes 38%, while Topic 3, emphasizing products with natural ingredients, makes up 13.5% of the reviews. These findings can assist businesses in developing products that align more closely with consumer preferences. Moreover, they support prospective buyers in making informed purchasing decisions by enhancing their understanding of product attributes based on user experiencesToday, beauty products are essential especially for women. Because they\u27re so popular, review sites now offer information on products. These internet reviews include views on trends, customer satisfaction and product performance. It\u27s hard to draw conclusions from so many reviews. Large amounts of text have been analyzed using topic modelling techniques like Latent Dirichlet Allocation (LDA). LDA is a probabilistic model that explains data and why some sections are similar. LDA is a useful tool for text mining and information retrieval. Recent studies show that topic modelling for product reviews in the cosmetics business can provide useful insights into consumer perceptions and product qualities. The purpose of this study is to examine themes in customer evaluations of ten different brands of face wash products from the female daily website. Data collection, preprocessing, topic modeling using LDA, visualization, and interpretation of topics are all steps in the research procedure. The findings show that Topic 2, which captures user conversations about product benefits that users prefer, is the most frequently discussed topic by users when evaluating a product (48.5%). This is followed by Topic 1, which captures user conversations about the effects of products on acne-prone skin (38%). Finally, Topic 3 captures user conversations about products based on natural ingredients (13.5%). These topics provide valuable insights for both manufacturers, who can improve product offerings, and consumers, helping them make informed purchasing decisions
Sistem Pendukung Keputusan Berbasis K-Means dan AHP Hybrid untuk Pemantauan Asrama
No full textPengelolaan asrama di Pusat Pengembangan Sumber Daya Manusia untuk Alat Transportasi (PPSDMAP) menghadapi tantangan dalam alokasi kamar, pemantauan fasilitas, dan pengambilan keputusan akibat ketergantungan pada sistem manual. Studi ini bertujuan untuk mengembangkan sistem pendukung keputusan cerdas (DSS) berdasarkan integrasi K-Means Clustering dan Proses Hierarki Analitis (AHP) guna meningkatkan efisiensi pengelolaan asrama. Pendekatan kuantitatif diterapkan dengan metode pengembangan sistem yang meliputi tahap perencanaan, pengumpulan data, analisis, desain, implementasi, dan pengujian. Data diperoleh melalui observasi dan wawancara di PPSDMAP, kemudian diproses menggunakan K-Means untuk pengelompokan data kamar asrama dan AHP untuk menentukan prioritas perbaikan fasilitas. Hasil menunjukkan bahwa sistem berhasil mengelompokkan data menjadi tiga kluster dengan nilai validitas Davies Bouldin Index sebesar 0.52, serta menghasilkan keputusan prioritas berdasarkan kriteria layanan, fasilitas, dan keamanan. Alternatif terbaik yang diidentifikasi adalah peningkatan konektivitas internet dengan bobot 49,25%, diikuti oleh pemasangan CCTV dan layanan laundry. Sistem ini secara efektif mendukung manajer dalam mengambil keputusan yang lebih cepat dan akurat berdasarkan data, dengan implikasi praktis untuk efisiensi manajemen fasilitas dan peningkatan kenyamanan penghuni asrama.Dormitory management at the Human Resources Development Center for Transportation Apparatus (PPSDMAP) encounters challenges in room allocation, facility monitoring, and decision-making due to reliance on manual systems. This study develops an intelligent decision support system (DSS) that integrates K-Means clustering and the Analytical Hierarchy Process (AHP) to improve management efficiency. A quantitative methodology was employed, encompassing planning, data collection, analysis, design, implementation, and testing. Data collected through observation and interviews at PPSDMAP were processed using K-Means for clustering dormitory room data and AHP for prioritizing facility improvements. The results show that the system successfully grouped data into three clusters, achieving a Davies Bouldin Index validity value of 0.52, and generated priority decisions based on service, facility, and security criteria. Improving internet connectivity was identified as the top alternative with a weight of 49.25%, followed by CCTV installation and laundry services. The system enables managers to make faster and more accurate data-driven decisions, offering practical benefits for facility management efficiency and enhancing the comfort of dormitory occupants
Predictive Modeling of Wildlife Trade Using the k-Nearest Neighbor Algorithm
Full text linkPerdagangan satwa liar dengan pengawasan yang tidak optimal dapat merupakan salah satu ancaman yang mampu memberikan dampak signifikan bagi keberlangsungan keanekaragaman hayati. Beberapa spesies hewan seperti mamalia dan reptil saat ini terancam punah dengan adanya perdagangan satwa liar yang tidak terpantau dengan baik. Praktik perdagangan satwa liar kerap ditemukan di berbagai media, termasuk situs internet, karena kemudahan akses yang ditawarkan kepada masyarakat luas. Namun jika data tersebut diolah secara manual maka dibutuhkan waktu, tenaga, dan upaya yang cukup besar. Oleh sebab itu, dibutuhkan sebuah pendekatan berbasis kecerdasan buatan untuk mengidentifikasi wilayah dengan aktivitas perdangan satwa liar yang tinggi dan berpotensi melampai batas perdagangan satwa liar yang diizinkan. Penelitian ini bertujuan untuk membangun sebuah model yang dapat melakukan prediksi area yang rawan perdagangan satwa liar. Algoritma yang digunakan dalam penelitian ini adalah algoritma clustering K-Means untuk mengelompokkan wilayah berdasarkan tingkat kerawanan aktivitas perdagangan satwa liar, serta algoritma k-Nearest Neighbor untuk melakukan prediksi tingkat kerawanan wilayah perdagangan satwa liar. Selain itu, metode Market Basket Analysis digunakan untuk mengidentifikasi pola asosiasi dalam perdagangan satwa liar antar negara. Data yang digunakan merupakan data perdagangan satwa liar dari berbagai negara pada tahun 2018 hingga 2020. Dengan menggunakan pendekatan clustering, dalam penelitian ini diklasifikasikan tiga wilayah dengan potensi kerawanan perdagangan satwa liar, yaitu wilayah dengan resiko rendah, sedang, dan tinggi. Hasil penelitian menunjukkan bahwa model prediksi yang dibangun mampu memprediksi wilayah rawan perdagangan satwa liar dengan tingkat akurasi model training sebesar 99% dengan data impor dan 100% dengan data ekspor. Setelah dievaluasi dengan 3-cross fold validation, akurasi model yang diperoleh adalah sebesar 97% untuk data impor dan 98% untuk data ekspor. Hasil akurasi model testing dalam penelitian ini adalah sebesar 100% dengan data impor maupun ekspor. Melalui pendekatan metode market basket analysis, penelitian ini menyimpulkan dengan data yang dipertimbangkan belum ditemukan pola asosiasi yang kuat dalam aktivitas perdagangan satwa liar antara satu negara spesifik dengan negara lainnya.Wildlife trade conducted under suboptimal supervision can pose a significant threat to the long-term sustainability of biodiversity. Several animal species, including mammals and reptiles, are now facing extinction due to the lack of proper monitoring of wildlife trade. Wildlife trade practices are frequently found across various media platforms, including websites, due to the ease of access they offer to the public. However, if the data is processed manually, it requires considerable time, effort, and resources. To address this issue, an artificial intelligence-based approach is needed to help process wildlife trade data so that areas vulnerable to wildlife trafficking can be predicted quickly and accurately. Therefore, an artificial intelligence-based approach is needed to identify regions with high wildlife trade activity that are at risk of exceeding the permitted trade limits. This study aims to develop a model capable of predicting regions at risk of wildlife trade. The algorithms used in this study are the K-Means clustering algorithm to classify regions based on the risk level of wildlife trade areas, and the k-Nearest Neighbor algorithm to predict the risk level of wildlife trade areas. In addition, the Market Basket Analysis method is used to identify association patterns in wildlife trade between countries. The data used consists of wildlife trade data from various countries from 2018 to 2020. Using the clustering approach, this study classifies three levels of risk for wildlife trafficking: low, medium, and high. The results of the study show that the predictive model developed is capable of identifying areas vulnerable to wildlife trade, achieving a training accuracy of 99% with import data and 100% with export data. After being evaluated using 3-fold cross-validation, the model achieved an accuracy of 97% for import data and 98% for export data. The testing accuracy obtained in this study was 100% for both import and export data. Through the market basket analysis approach, this study concludes that, based on the data considered, no strong association patterns have been found in wildlife trade activities between specific countries
Enhancing Usability of the Qualitiva Educational Applications: A Mixed-Methods Study using SUS and Heuristic Evaluation
Full text linkStudi ini mengevaluasi kegunaan aplikasi Qualitiva menggunakan metode System Usability Scale (SUS) dan Heuristic Evaluation (HE). Tujuan dari penelitian ini adalah untuk mengidentifikasi kekuatan dan kelemahan aplikasi, serta memberikan rekomendasi perbaikan. Pendekatan kuantitatif digunakan dengan mendistribusikan kuesioner SUS kepada 100 responden, sementara pendekatan kualitatif melibatkan evaluasi heuristik oleh dua ahli usability. Hasil menunjukkan skor rata-rata SUS sebesar 69,00 (Grade C), yang menunjukkan penerimaan pengguna secara umum namun menyoroti beberapa area yang perlu perbaikan dalam kecepatan, desain, keamanan data, dan efisiensi sistem. Evaluasi heuristik mengidentifikasi masalah seperti kejelasan status sistem, penggunaan terminologi yang ramah pengguna, desain antarmuka, dan fitur pencegahan kesalahan. Penelitian ini meningkatkan pemahaman tentang kegunaan dalam aplikasi pendidikan dan memberikan rekomendasi yang dapat ditindaklanjuti untuk meningkatkan pengalaman pengguna, termasuk mengoptimalkan kecepatan, memperbaiki antarmuka, dan meningkatkan panduan pengguna.This study evaluates the usability of the Qualitiva application using the System Usability Scale (SUS) and Heuristic Evaluation (HE) methods. The goal is to identify the application’s strengths and weaknesses, offering recommendations for improvement. A quantitative approach was used with SUS questionnaires distributed to 100 respondents, while a qualitative approach involved heuristic evaluation by two usability experts. Results show an average SUS score of 69.00 (Grade C), indicating general user acceptance but highlighting areas for improvement in speed, design, data security, and system efficiency. The heuristic evaluation identified issues such as system status clarity, user-friendly terminology, interface design, and error prevention features. This research enhances understanding of usability in educational applications and provides actionable recommendations to improve user experience, including optimizing speed, refining the interface, and enhancing user guidance
Analisis Pembiayaan Sistem Digital Signage dengan Metode Function Point Analysis pada Industri F&B
Full text linkDigital Signage merupakan teknologi strategis dalam industri makanan dan minuman (F&B) yang memungkinkan penyampaian konten dinamis, seperti menu, promosi, dan iklan secara real-time, sehingga meningkatkan pengalaman pelanggan dan efisiensi operasional. Sistem Digital Signage pada penelitian ini dirancang menggunakan arsitektur microservices berbasis Java Spring Boot dan React JS, serta terintegrasi dengan platform MagicInfo untuk mendukung manajemen konten secara efisien. Fitur Sync Play memungkinkan penayangan konten secara sinkron di berbagai layar. Penelitian ini bertujuan untuk mengukur fungsionalitas dan kompleksitas sistem menggunakan metode Function Point Analysis (FPA), guna menghasilkan estimasi yang akurat terkait upaya, biaya, dan waktu pengembangan. Hasil perhitungan menunjukkan selisih antara estimasi dan ukuran aktual sebesar 1,68%, yang mencerminkan tingkat akurasi tinggi dari metode ini. Penelitian ini diharapkan dapat mendukung transformasi digital di industri F&B melalui implementasi solusi Digital Signage yang inovatif dan efektif.Digital signage has emerged as a strategic technology in the food and beverage (F&B) industry, delivering dynamic content—such as menus, promotions, and advertisements—in real-time to enhance customer experience and operational efficiency. This study presents a digital signage system designed using a microservices architecture based on Java Spring Boot and React JS, integrated with the MagicInfo platform to facilitate efficient content management. The Sync Play feature enables synchronized content display across multiple screens. The functionality and complexity of the system are evaluated using the Function Point Analysis (FPA) method to generate accurate estimates of development effort, cost, and time. The results indicate a deviation of only 1.68% between the estimated and actual system size, demonstrating the high accuracy of the method. These findings are expected to support digital transformation in the F&B industry through the deployment of innovative and effective Digital Signage solutions
Evaluasi Modular UX terhadap Fitur Narasi Budaya pada Sistem Rekomendasi Makanan
No full textThis study designs and evaluates a cultural narrative feature in a food recommender system to enhance emotional engagement and cultural relevance while preserving usability. The feature uses progressive disclosure and is implemented as a modular interface component that embeds local food stories within recommendation cards. Fifteen users participated in a task-based, triangulated UX evaluation (usability testing, heuristic evaluation, and thematic analysis). Although only 25% of participants interacted with the narrative element, those who did spent more than twice as long engaging with the content. Heuristic findings indicated issues in system-status visibility and visual transition, while thematic insights showed curiosity, emotional recall, and cultural identification. The results suggest that optional, contextual narratives can enrich recommender-system UX and can be integrated as maintainable modular components for iterative refinementPenelitian ini menyajikan perancangan dan evaluasi user experience (UX) terhadap fitur cultural narrative dalam sistem rekomendasi makanan, dengan tujuan meningkatkan keterlibatan emosional dan relevansi budaya tanpa mengorbankan aspek usability. Fitur tersebut dirancang menggunakan pendekatan progressive disclosure dan diimplementasikan sebagai komponen antarmuka modular yang menyisipkan kisah makanan lokal di dalam kartu rekomendasi. Sebanyak lima belas pengguna berpartisipasi dalam evaluasi berbasis tugas menggunakan metode triangulatif: usability testing, heuristic analysis, dan thematic analysis. Hasil menunjukkan bahwa meskipun hanya 25 persen pengguna mengklik elemen narasi, mereka yang melakukannya menghabiskan waktu lebih dari dua kali lipat dalam berinteraksi dengan konten tersebut. Evaluasi heuristik mengungkapkan masalah pada aspek visibilitas dan transisi visual, sementara analisis tematik mengonfirmasi bahwa fitur ini membangkitkan rasa ingin tahu, ingatan emosional, dan identifikasi budaya. Temuan ini menunjukkan bahwa contextual narrative dapat memperkaya pengalaman pengguna secara bermakna dalam sistem rekomendasi. Dengan mengadopsi struktur UX modular, penelitian ini juga mengilustrasikan bagaimana fitur kontekstual dapat disisipkan dan disempurnakan secara iteratif dalam antarmuka rekomendasi digital, serta mendukung desain sistem yang berskala dan mudah dipelihara
Analisis Spasial-temporal Titik Panas dan PM2.5 di Riau, Jambi, dan Sumatera Selatan
Full text linkKebakaran hutan dan lahan (karhutla) di Indonesia berkontribusi signifikan terhadap penurunan kualitas udara melalui peningkatan konsentrasi PM2.5. Penelitian ini menganalisis pola spasial-temporal sebaran titik panas dan estimasi konsentrasi PM2.5 di Provinsi Riau, Jambi, dan Sumatera Selatan selama Agustus–Oktober 2023. Data titik panas MODIS dianalisis menggunakan algoritma ST-DBSCAN dengan pengaturan parameter jarak spasial, jarak temporal, dan jumlah minimum titik untuk mengidentifikasi klaster kebakaran. Estimasi PM2.5 diperoleh dari konversi Aerosol Optical Depth (AOD) MODIS menggunakan model empiris. Hasil menunjukkan bahwa ST-DBSCAN efektif dalam mengidentifikasi klaster titik panas, dengan kepadatan klaster tertinggi teramati di Provinsi Sumatera Selatan. Rata-rata estimasi PM2.5 tercatat sebesar 50,51 µg/m³ di Provinsi Riau, 48,16 µg/m³ di Provinsi Jambi, dan 41,59 µg/m³ di Provinsi Sumatera Selatan. Konsentrasi PM2.5 tertinggi terjadi di Provinsi Riau pada bulan Oktober dan melampaui ambang batas pedoman kualitas udara WHO. Temuan ini menegaskan adanya keterkaitan kuat antara dinamika spasial-temporal karhutla dan peningkatan polusi udara, serta menunjukkan potensi pendekatan ini dalam mendukung analisis risiko lingkungan dan kesehatan.Forest and land fires in Indonesia significantly degrade air quality by increasing PM2.5 concentrations. This study examines the spatiotemporal patterns of fire hotspot distribution and estimated PM2.5 concentrations in Riau, Jambi, and South Sumatra Provinces during August–October 2023. MODIS fire hotspot data were analyzed using the ST-DBSCAN algorithm with defined spatial distance, temporal distance, and minimum point parameters to identify fire clusters. PM2.5 concentrations were estimated by converting MODIS Aerosol Optical Depth (AOD) using an empirical model. The results demonstrate that ST-DBSCAN effectively identifies fire hotspot clusters, with the highest cluster density observed in South Sumatra Province. The average estimated PM2.5 concentrations were 50.51 µg/m³ in Riau, 48.16 µg/m³ in Jambi, and 41.59 µg/m³ in South Sumatra. The highest PM2.5 levels occurred in Riau Province in October, exceeding the World Health Organization air quality guideline. These findings reveal a strong spatiotemporal association between fire activity and elevated particulate pollution and highlight the potential of this approach to support environmental and health risk assessments related to wildfire events
Augmentasi Data Sintetis Berbasis DCGAN untuk Citra Neoplasia Intraepitel Serviks (CIN)
No full textOne of the major challenges in medical imaging is the limited availability of high-quality datasets. To address this, Generative Artificial Intelligence (Generative AI) offers a promising solution by generating synthetic medical images to augment existing datasets. This study explores the application of Deep Convolutional Generative Adversarial Networks (DCGAN) for data augmentation in CIN imaging. Two training scenarios were implemented: DCGAN with manual data augmentation and another without manual augmentation. The image quality was evaluated using the Fréchet Inception Distance (FID). The results indicate that incorporating data augmentation improves the stability of training and enhances the quality of generated images FID scores of 2.21. In contrast, training DCGAN without manual augmentation resulted in a higher FID score of 2.52, indicating lower image quality. These findings highlight the effectiveness of DCGAN in medical image augmentation and its potential to enhance deep learning-based diagnostic models for cervical cancer detection or classification.Salah satu tantangan utama dalam pencitraan medis adalah keterbatasan ketersediaan dataset berkualitas tinggi. Untuk mengatasi hal ini, Generative Artificial Intelligence (AI Generatif) menawarkan solusi menjanjikan dengan menghasilkan citra medis sintetis guna memperluas dataset yang ada. Penelitian ini mengeksplorasi penerapan Deep Convolutional Generative Adversarial Networks (DCGAN) untuk augmentasi data pada citra CIN (Cervical Intraepithelial Neoplasia). Dua skenario pelatihan diterapkan: DCGAN dengan augmentasi data manual dan tanpa augmentasi manual. Kualitas citra dievaluasi menggunakan Fréchet Inception Distance (FID). Hasil penelitian menunjukkan bahwa penerapan augmentasi data meningkatkan stabilitas pelatihan dan memperbaiki kualitas citra yang dihasilkan dengan skor FID sebesar 2,21. Sebaliknya, pelatihan DCGAN tanpa augmentasi manual menghasilkan skor FID lebih tinggi, yaitu 2,52, yang menunjukkan kualitas citra lebih rendah. Temuan ini menegaskan efektivitas DCGAN dalam augmentasi citra medis serta potensinya untuk meningkatkan model diagnosis berbasis deep learning pada deteksi atau klasifikasi kanker serviks