Jurnal Buana Informatika

Penerapan Optical Character Recognition untuk Pengenalan Variasi Teks pada Media Presentasi Pembelajaran

Kristian Adi Nugraha — Mon, 01 Apr 2024 00:00:00 +0700

Media digital merupakan bentuk utama media pembelajaran yang banyak digunakan untuk kegiatan belajar mengajar di kelas saat ini. Media pembelajaran digital umumnya tersimpan dalam bentuk citra karena memiliki unsur visual di dalamnya. Salah satu kelemahan data dalam bentuk citra adalah seluruh isi di dalamnya dianggap sebagai gambar, sementara pada media pembelajaran juga terdapat unsur teks di dalamnya. Oleh karena itu, dibutuhkan metode OCR untuk membaca teks di dalamnya agar media tersebut dapat diolah lebih lanjut, misalnya untuk keperluan kategorisasi (indexing) atau untuk dibaca pada sistem lain seperti chatbot. Umumnya, metode OCR digunakan untuk mengenali tulisan dengan bentuk yang seragam pada sebuah citra. Sedangkan pada media pembelajaran, teks di dalamnya memiliki variasi yang berbeda-beda. Penelitian ini mencoba menerapkan metode OCR dengan menggunakan Tesseract untuk menguji 30 data media pembelajaran yang memiliki berbagai macam variasi teks dalam sebuah citra. Hasil pengujian menunjukkan tingkat akurasi pengenalan teks yang cukup baik, yaitu sebesar 91,11%.

Gamified Distance Learning Application Design for Enhanced Student Engagement and User Experience

Fedelis Brian Putra Prakasa, Joseph Eric Samodra, Thomas Adi Purnomo Sidhi — Mon, 01 Apr 2024 00:00:00 +0700

Distance Learning in Indonesia is one of the learning methods that began to be applied during the Covid-19 pandemic. Yet students face some obstacles, such as lack of motivation, struggling with operating learning devices, difficulty maintaining focus, and student engagement during the learning process. Gamification offers a solution to these problems by significantly enhancing user motivation and engagement, as it has been tested in research to have a profound impact. Therefore, this study aims to design a mobile application for Distance Learning by implementing gamification. It employs qualitative and quantitative data, including 32 students' responses from questionnaires like UEQ-S, utilized for testing user interface, and UES-SF, employed for testing gamification elements. By implementing gamification in this design, an engagement score of 83% was obtained, and the overall UEQ-S result was 1.89 in the Excellent category.

Male Fertility Classification using Machine Learning and Oversampling Techniques

Aloysius Gonzaga Pradnya Sidhawara — Mon, 01 Apr 2024 00:00:00 +0700

Machine learning methods have been applied to male fertility diagnosis in recent years. Through early infertility case detection, this technology application offers potential benefits to the medical field. This study presents an experimental investigation that examines the prospect of using the oversampling technique and feature selection to enhance the performance of shallow classifiers to classify male fertility on the Fertility Dataset. Two oversampling techniques (SMOTE and ADASYN), two different scalers (MinMax and Standard), and two different feature selection methods (SelectKBest and SelectFromModel) were used to improve the performance of the classifier. The results show that the performance of machine learning models is better on the oversampled dataset than the original dataset. Random Forest performed best on the SMOTE test set with 90% accuracy, 89% and 100% Recall in Normal and Altered classes, respectively. Accidents or trauma, Age, and High Fevers features are selected by SelectKBest, and considered as factors that contribute to male fertility in prior studies.

Machine Learning for Environmental Health: Optimizing ConcaveLSTM for Air Quality Prediction

Mon, 01 Apr 2024 00:00:00 +0700

This study investigates the optimization of the ConcaveLSTM model for air quality prediction, focusing on the interplay between input sequence lengths and the number of LSTM units to enhance forecasting accuracy. Through the evaluation of various model configurations against performance metrics such as RMSE, MAE, MAPE, and R-squared, an optimal setup featuring 50 input steps and 300 neurons was identified, demonstrating superior predictive capabilities. The findings underscore the critical role of model parameter tuning in capturing temporal dependencies within environmental data. Despite limitations related to dataset representativeness and environmental variability, the research provides a solid foundation for future advancements in predictive environmental modeling. Recommendations include expanding dataset diversity, exploring hybrid models, and implementing real-time data integration to improve model generalizability and applicability in real-world scenarios.

Konfigurasi Model Prophet Untuk Prediksi Harga Saham Sektor Teknologi di Indonesia Yang Akurat

Ravelino Sebastian Santoso, Findra Kartika Sari Dewi — Mon, 01 Apr 2024 00:00:00 +0700

Saham merupakan salah satu instrumen investasi yang sedang ramai dan digemari oleh masyarakat muda Indonesia. Untuk dapat meramal harga saham, dapat dilakukan analisis teknikal dengan menerapkan machine learning. Namun, untuk dapat menggunakan machine learning, diperlukan implementasi algoritma yang membutuhkan waktu panjang serta keterampilan tinggi. Maka dari itu digunakanlah model Prophet, model machine learning yang mudah untuk dikembangkan. Pengembangan dilakukan dengan menyesuaikan karakteristik data saham yang merupakan data bertipe time series. Eksperimen dilakukan untuk menemukan konfigurasi yang perlu dilakukan terhadap model dalam menghasilkan peramalan yang paling akurat. Melalui penelitian yang dilakukan, hasil terbaik yang didapatkan adalah model Prophet yang menggunakan dataset paling banyak dan juga melalui hyperparameter tuning. Hal ini dapat dibuktikan dengan visualisasi yang ada serta nilai error yang rendah, dimana MAPE (Mean Absolute Percentage Error) mempunyai nilai error sebesar 15%.