Claim Missing Document
Check
Articles

Found 1 Documents
Search
Journal : Journal of Robotics and Control (JRC)

HyVADSVM: Hybrid VADER-SVM and GridSearchCV Optimization for Enhancing Cyberbullying Detection Ernawati, Siti; Frieyadie, Frieyadie; Yulia, Eka Rini
Journal of Robotics and Control (JRC) Vol. 6 No. 1 (2025)
Publisher : Universitas Muhammadiyah Yogyakarta

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.18196/jrc.v6i1.24385

Abstract

Cyberbullying detection is becoming increasingly crucial in today’s digital era, as many individuals suffer from online harassment. The main challenge lies in accurately identifying patterns of harassment in social media texts, which often use informal languages, slang, and sarcasm. Existing methods struggle to capture emotional context owing to the vast amount of data and rapid digital interactions. This study aims to improve the detection accuracy by combining advanced sentiment analysis using VADER and parameter tuning with GridSearchCV. Data were collected from Instagram, Twitter, and YouTube, with TF-IDF employed for feature extraction. Multiple machine-learning classifiers (SVM, K-NN, NB, LR, DT, and RF) were tested to determine the best-performing model. VADER was selected for its reliability in processing social media texts rich in informal contexts, effectively capturing emotional nuances, such as sarcasm and varying sentiment intensities. This makes it well suited for complex language patterns typical of cyberbullying scenarios, enhancing data labeling and analysis accuracy. Using 10-fold cross-validation for reliable testing, performance metrics (accuracy, precision, recall, and F1-Score) were evaluated using a confusion matrix. The findings highlight SVM as the most effective model when optimized with GridSearchCV, achieving accuracy (98.83%), precision (98.78%), recall (98.83%), and F1-Score (98.62%) with kernel =linear, C=1, and gamma=scale. This optimized model, HyVADSVM model has significant potential in cyberbullying detection, contributing to academic research and serving as an effective tool to prevent online harassment. Future work could integrate this model into real-time systems, improve user safety, and support digital policymaking.