30 research outputs found
Crawling Web berdasarkan Ontology
No full textTelah dikembangkan sebuah aplikasi Web Crawler untuk melakukan penjelajahan dan pengambilan
halaman-halaman Web yang ada di Internet. Program crawl memanfaatkan WordNet dan ontology dari
struktur Open Directory Project (ODP) untuk mencari relevansi suatu halaman web dengan kata kunci.
Pentingnya suatu halaman web dihitung menggunakan rumus similaritas tekstual. Pengujian dilakukan
untuk membandingkan harvest-rate crawling menggunakan WordNet dengan menggunakan ontology dari
struktur ODP
COMBINATION K-MEANS AND LSTM FOR SOCIAL MEDIA BLACK CAMPAIGN DETECTION OF INDONESIA PRESIDENTIAL CANDIDATES 2024
No full textSocial media has become the main platform for the public and political figures to voice opinions and run political campaigns. Despite its positive impact, social media also has negative impacts, particularly in the spread of Black Campaigns. This phenomenon has become critical, especially about the 2024 elections in Indonesia that target presidential candidates. Black campaigns can trigger conflict and damage the image of presidential candidates in the eyes of the public. Therefore, it is important to detect black campaigns against presidential candidates. This research develops a Black Campaign detection model using the K-means clustering algorithm and the Long Short-Term Memory (LSTM) approach. K-means is implemented to cluster text data on Twitter social media, while LSTM is used to learn word order patterns and detect text. The result is that K-means can effectively prepare the data, and classification using LSTM shows an accuracy of 90.28%. The comparison with Ensemble Learning classification model achieved an accuracy of 94.31%. Evaluation involved accuracy, precision, recall, and F1-score, with the result that Ensemble Learning was slightly superior in the evaluation matrix. However, compared to Ensemble Learning, LSTM has an advantage in understanding word order, which can be achieved by utilizing the advantages of Deep Learning Recurrent Neural Network architecture. Testing on sample data shows the similarity between LSTM and Ensemble Learning models in detecting Black Campaigns on Twitter social media post text data
Time Series Forecasting for Container Throughput Using SARIMA and LSTM: A Case Study of Tanjung Emas Port, Semarang
No full textAbstract: Accurate forecasting of container throughput is vital for enhancing strategic planning and operational efficiency in seaport management. This study compares the performance of two time series forecasting models Seasonal Autoregressive Integrated Moving Average (SARIMA) and Long Short-Term Memory (LSTM) in predicting container throughput at Tanjung Emas Port, Semarang, Indonesia. Monthly throughput data from January 2014 to April 2025 were preprocessed using stationarity transformation and normalization techniques. Model performance was evaluated using Mean Absolute Percentage Error (MAPE), Root Mean Squared Error (RMSE), and Mean Absolute Error (MAE). The SARIMA model effectively captured seasonal patterns and produced accurate short-term forecasts. Conversely, the LSTM model exhibited notable significant deviation from the actual data , indicating lower predictive performance in this context. The findings indicate that SARIMA currently provides a more reliable forecasting approach for the port. Future research should consider hybrid models (e.g., SARIMA-LSTM) and incorporate exogenous variables to improve forecasting accuracy and support data-driven decision-making in port operations
Sistem Rekomendasi Pemilihan Mobil Menggunakan Metode Edas
No full textBerdasarkan data Gabungan Industri Kendaraan Bermotor Indonesia atau GAIKINDO, penjualan mobil mengalami penurunan pada tahun 2020 dan 2021 akibat wabah Covid-19, namun meningkat sebesar 49% pada tahun 2022 dan 2023. Hal ini menunjukkan bahwa penjualan kendaraan pribadi , mobil lebih disukai masyarakat Indonesia dibandingkan pilihan alternatif. Untuk memilih mobil berdasarkan merek, model, tipe, harga, kapasitas mesin, transmisi, tahun pembuatan, dan kapasitas bensin, penelitian ini menggunakan pendekatan EDAS. Mobil Honda CRV dengan nilai AS_i = 0,812 merupakan rekomendasi yang muncul dari kriteria pemilihan merek Honda dan girboks CVT, dan mobil Honda Brio dengan nilai AS_i= 0,713 merupakan rekomendasi kedua dan mobil Honda BRV dengan nilai AS_i= 0,713 merupakan rekomendasi kedua. nilai AS_i= 0,6888 merupakan rekomendasi ketiga
Interpretasi model Stacking Ensemble untuk analisis sentimen ulasan aplikasi pinjaman online menggunakan LIME
No full textLocal Interpretable Model-agnostic Explanations (LIME) can be used to overcome black box problems in the results of sentiment analysis classification models. This research uses reviews of online loan applications on the Play Store as a dataset. Each classification model has weaknesses and its performance can be improved by using stacking ensembles, especially to overcome the problem of imbalanced data classes. The dataset that has been obtained will be cleaned, pre-processed and converted into a numerical vector using TF-IDF. Classification is carried out using three basic models, namely random forest, naïve Bayes and support vector machine (SVM). The output of the basic classification model is used as an input for stacking ensemble logistic regression. Based on the comparison of the four models, stacking ensemble has the best performance with an accuracy of 87.05%. The application of LIME for interpreting classification models with sample data succeeded in explaining the factors that influence model decisions with a prediction probability of 95% and in accordance with manual observations. The results of this research can be used as insight and education to the public about the ease of online loan and its dangers, which are reflected in the positive and negative sentiments in a review.Local Interpretable Model-agnostic Explanations(LIME) dapat digunakan untuk mengatasi masalah blackbox pada hasil model klasifikasi analisis sentimen. Penelitian ini menggunakan ulasan aplikasi pinjaman online di play store sebagai dataset. Masing-masing model klasifikasi memiliki kelemahan dan dapat ditingkatkan kinerjanya dengan menggunakan stacking ensemble terutama untuk mengatasi permasalahan kelas data yang tidak seimbang. Dataset yang sudah diperoleh, dilakukan pembersihan data, pre-processing serta dirubah menjadi vektor numerik menggunakan TF-IDF. Klasifikasi dilakukan dengan tiga model dasar yaitu random forest, naïve bayes dan support vector machine(SVM). Luaran dari model klasifikasi dasar dijadikan sebagai masukan bagi stacking ensemble logistic regression. Berdasarkan komparasi keempat model, stacking ensemble memiliki kinerja terbaik dengan akurasi 87,05%. Penerapan LIME untuk intrepretasi model klasifikasi dengan sampel data berhasil menjelaskan faktor-faktor yang berpengaruh terhadap keputusan model dengan probabilitas prediksi 95% dan sesuai dengan pengamatan manual. Hasil penelitian ini bisa digunakan sebagai wawasan dan edukasi kepada masyarakat tentang kemudahan pinjol dan bahayanya yang tercermin dari sentimen positif dan negatif pada sebuah ulasan
IMPLEMENTASI METODE DESIGN THINKING PADA PERANCANGAN WEBSITE PROFIL DEPARTEMEN TEKNIK KOMPUTER
No full textPenelitian ini mengimplementasikan Metode Design Thinking dalam pengembangan website profil Departemen Teknik Komputer untuk meningkatkan pengalaman pengguna dan aksesibilitas. Studi ini menjawab tantangan pembuatan antarmuka yang intuitif dan mudah digunakan bagi pemangku kepentingan akademik. Dengan pendekatan mixed-method, penelitian mengikuti lima tahap design thinking: empati, definisi, ideasi, prototipe, dan pengujian. Data dikumpulkan melalui wawancara mendalam dengan 15 stakeholder, pengujian usability, dan evaluasi System Usability Scale (SUS) melibatkan 30 responden. Hasil penelitian menunjukkan peningkatan signifikan dalam usability dengan skor SUS rata-rata 78,4 (kategori "baik"), membuktikan keberhasilan penerapan prinsip user-centered design. Temuan ini menegaskan efektivitas design thinking dalam menyelesaikan masalah kompleks antarmuka di lingkungan akademik. Penelitian berkontribusi secara teoretis dengan memperkaya literatur tentang design thinking dalam pengembangan website pendidikan, serta secara praktis dengan menyediakan kerangka kerja yang dapat diadopsi institusi serupa. Rekomendasi mencakup pengujian iteratif dan pelibatan stakeholder untuk solusi digital berkelanjutan.
Kata Kunci: Design Thinking, pengembangan website, usability, profil akademik, desain berpusat penggun
KLASIFIKASI TEKNIK BULUTANGKIS BERDASARKAN POSE DENGAN CONVULUTIONAL NEURAL NETWORK
No full textConvolutional Neural Network (CNN) is a deep learning algorithm which is the development of Multilayer Perception (MLP) which is designed to process data in two-dimensional form. At the stage of making the system there are several stages including sample data, data sources and data analysis methods. The dataset that is processed is the Badminton Technique, namely the Forehand Technique which consists of 374 images, the Service Technique consisting of 369 images and the smash technique consisting of 420 images with outliers of 146 images. After the data is cleaned of outliers, bootrapping is carried out again to unite all data from each separate class into one again. The results of this study say that the Classification of Badminton Techniques Based on Pose with Convolutional Neural Networks, it can be concluded that the process of testing pose classification with test data using several methods such as logistic regression, random forest, and KNN produces significant accuracy. values ranging from 80% to 90%.Convolutional Neural Network (CNN) is a deep learning algorithm which is the development of Multilayer Perception (MLP) which is designed to process data in two-dimensional form. At the stage of making the system there are several stages including sample data, data sources and data analysis methods. The dataset that is processed is the Badminton Technique, namely the Forehand Technique which consists of 374 images, the Service Technique consisting of 369 images and the smash technique consisting of 420 images with outliers of 146 images. After the data is cleaned of outliers, bootrapping is carried out again to unite all data from each separate class into one again. The results of this study say that the Classification of Badminton Techniques Based on Pose with Convolutional Neural Networks, it can be concluded that the process of testing pose classification with test data using several methods such as logistic regression, random forest, and KNN produces significant accuracy. values ranging from 80% to 90%
Integrasi Algoritma Apriori dan K-Means untuk Optimalisasi serta Analisis Pola Pemasaran Suku Cadang Otomotif
No full textThe automotive parts industry in Indonesia faces challenges of inefficient inventory management and a lack of understanding of customer purchasing patterns, resulting in overstocking or stockouts that cause financial losses. This study aims to apply association rule mining and K-Means clustering to automotive parts retail transaction data to uncover purchasing patterns and product segmentation to support effective inventory management and marketing strategies. Our research is quantitative in nature with a dataset of 14,165 transactions analyzed using Google Colab-Python. The Apriori algorithm was applied with a minimum support of 1% and confidence of 50%, while K-Means clustering was used for product segmentation with normalized numerical attributes. Association rule mining identified 15 significant rules with the strongest pattern between differential oil and brake fluid (confidence 73.5%, lift ratio 5.515). K-Means produced seven optimal clusters (silhouette score 0.68) that categorized products into premium, fast-moving, slow-moving, and other specific characteristics. The main contribution is an integrated framework that combines clustering to enrich the interpretation of association rules, enabling effective bundling strategies. Practical implications include a cross-selling recommendation system (15-20% revenue increase), differential pricing per cluster, and stock predictions that reduce overstocking by 25% and avoid stock-outs, supporting the digital transformation of data-driven retail management
Implementasi Metode YOLO pada Deteksi Pakaian Keselamatan yang Lengkap di Proyek Kontruksi
No full textSafety at work is important, so the use of personal protective equipment (ADP) is a must. However, in reality on the ground, there are relatively few workers who use complete and correct PPE. Due to this problem, the company as the person responsible employs K3 officers to monitor workers' use of PPE. To reduce the costs incurred in employing K3 officers, a system was created that was able to detect and monitor worker discipline in using PPE. Therefore, a personal protective equipment detection system was created. One method created to create object detection is the You Only Look Once (YOLO) method. The way YOLO works is by looking at the entire image once, then passing through the neural network once to directly detect existing objects. The results of this implementation aim to detect project workers who use complete protective equipment and do not use it, with the output results in the form of images that have been detected by people who use complete protective equipment or do not use it with labeling and bounding boxes on the detected image. From the test results on a total of 96 images, it shows that there is an accuracy value of 65%
