Optimization of the Activation Function for Predicting Inflation Levels to Increase Accuracy Values DOI Open Access
Agus Perdana Windarto,

Indra Riyana Rahadjeng,

Muhammad Noor Hasan Siregar

et al.

JURNAL MEDIA INFORMATIKA BUDIDARMA, Journal Year: 2024, Volume and Issue: 8(3), P. 1627 - 1627

Published: July 27, 2024

This study aims to optimize the backpropagation algorithm by evaluating various activation functions improve accuracy of inflation rate predictions. Utilizing historical data, neural network models were constructed and trained with Sigmoid, ReLU, TanH functions. Evaluation using Mean Squared Error (MSE) metric revealed that ReLU function provided most significant performance improvement. The findings indicate choice architecture significantly influences model's ability predict rates. In 5-7-1 architecture, Logsig demonstrated best performance, achieving lowest MSE (0.00923089) highest (75%) on test data. These results underscore importance selecting appropriate enhance prediction accuracy, outperforming other in context dataset used. research concludes optimizing is a crucial step developing more accurate models, contributing literature practical economic applications.

Language: Английский

Comparative Analysis of Classification Methods in Sentiment Analysis: The Impact of Feature Selection and Ensemble Techniques Optimization DOI Open Access
Sarjon Defit, Agus Perdana Windarto,

Putrama Alkhairi

et al.

Telematika, Journal Year: 2024, Volume and Issue: 17(1), P. 52 - 67

Published: Feb. 16, 2024

Optimizing classification methods (forward selection, backward elimination, and optimized selection) ensemble techniques (AdaBoost Bagging) are essential for accurate sentiment analysis, particularly in political contexts on social media. This research compares advanced models with standard ones (Decision Tree, Random Naive Bayes, Forest, K- NN, Neural Network, Generalized Linear Model), analyzing 1,200 tweets from December 10-11, 2023, focusing "Indonesia" "capres." It encompasses 490 positive, 355 negative, 353 neutral sentiments, reflecting diverse opinions presidential candidates issues. The enhanced model achieves 96.37% accuracy, the selection reaching 100% accuracy negative sentiments. study suggests further exploration of hybrid feature improved classifiers high-stakes analysis. With forward method, Bayes stands out classifying sentiments while maintaining high overall (96.37%).

Language: Английский

Citations

2
Optimization of the Activation Function for Predicting Inflation Levels to Increase Accuracy Values DOI Open Access
Agus Perdana Windarto,

Indra Riyana Rahadjeng,

Muhammad Noor Hasan Siregar

et al.

JURNAL MEDIA INFORMATIKA BUDIDARMA, Journal Year: 2024, Volume and Issue: 8(3), P. 1627 - 1627

Published: July 27, 2024

This study aims to optimize the backpropagation algorithm by evaluating various activation functions improve accuracy of inflation rate predictions. Utilizing historical data, neural network models were constructed and trained with Sigmoid, ReLU, TanH functions. Evaluation using Mean Squared Error (MSE) metric revealed that ReLU function provided most significant performance improvement. The findings indicate choice architecture significantly influences model's ability predict rates. In 5-7-1 architecture, Logsig demonstrated best performance, achieving lowest MSE (0.00923089) highest (75%) on test data. These results underscore importance selecting appropriate enhance prediction accuracy, outperforming other in context dataset used. research concludes optimizing is a crucial step developing more accurate models, contributing literature practical economic applications.

Language: Английский

Citations

0