Efendi, Akmar and Fitri, Iskandar and Nurcahyo, Gunadi Widi (2024) Improvement of Machine Learning Algorithms with Hyperparameter Tuning on Various Datasets. In: International Conference on Future Technologies for Smart Society (ICFTSS), 7-8 Agustus 2024, Kuala Lumpur, Malaysia.

Preview

Text 11. Prosidding Improvement of Machine Learning Algorithms with Hyperparameter Tuning.pdf - Published Version Download (19MB) | Preview

Abstract

In the digital era with a data explosion, classification techniques have become a primary aspect of machine learning, especially in Supervised Learning methods. These techniques allow computers to learn from existing data and apply their knowledge to classify new data based on patterns found in the training data. Although algorithms such as Support Vector Machine (SVM) and Naïve Bayes are reliable in many cases, they are not always optimal due to data complexity. This study evaluates the performance of various models and applies optimization techniques to enhance model performance across different datasets. The study uses three different datasets: academic data from the Faculty of Engineering at Universitas Islam Riau (UIR), tweet data from the social media platform X, and diabetes disease data from Kaggle. Each model is tested with a 70:30 data split, employing techniques such as SMOTE, Hyperparameter Optimization with Optuna, and XGBoost to improve model performance. The combination of SMOTE with SVM or GNB shows significant improvement in accuracy, precision, recall, and F1Score when optimization techniques are applied. For instance, the use of SMOTE, SVM, and Optuna achieves 100% accuracy on academic data, 97% on Twitter data, and 80% on diabetes data. Similarly, the combination of SMOTE, GNB, and XGBoost provides significant improvement. This study concludes that the application of optimization techniques like Optuna and the integration with algorithms such as XGBoost significantly enhance the performance of classification models across various datasets. This opens up opportunities for the development of more advanced and effective classification models in the future and makes a significant contribution to understanding the use of classification algorithms in various practical applications.

Actions (login required)

View Item