Evaluasi Algoritma Logistic Regression dan Random Forest pada Identifikasi Hate Speech

Adam Julizar; Mia Kamayani Sulaeman;

Evaluasi Algoritma Logistic Regression Dan Random Forest Pada Identifikasi

Hate Speech
Adam Julizar

Ujaran kebencian di media sosial merupakan ancaman serius terhadap kesejahteraan digital, terutama dalam
konteks kebahasaan yang beragam seperti Indonesia. Penelitian ini bertujuan untuk mengevaluasi dan
membandingkan kinerja dua algoritma machine learning, yaitu Logistic Regression dan Random Forest
dalam klasifikasi ujaran kebencian berbahasa Indonesia, serta mengkaji pengaruh penggunaan metode
ekstraksi fitur TF-IDF dan kombinasi TF-IDF + N-Gram terhadap performa klasifikasi. Penelitian ini
menggunakan dataset berbahasa Indonesia pada tahun 2019 yang terdiri dari 11.122 data yang diperoleh
dari platform Kaggle. Data dianalisis melalui beberapa tahapan, mulai dari praproses data (pembersihan

teks, normalisasi, penghapusan stopword, dan stemming), ekstraksi fitur menggunakan TF-IDF dan TF-
IDF + N-Gram, hingga klasifikasi menggunakan algoritma Logistic Regression dan Random Forest. Hasil

penelitian menunjukkan bahwa kombinasi TF-IDF + N-Gram dapat meningkatkan kinerja klasifikasi, di
mana akurasi Logistic Regression meningkat dari 81% menjadi 83%, dan Random Forest dari 82% menjadi
83%. Recall Logistic Regression juga meningkat dari 79% menjadi 85%. Logistic Regression unggul dari
segi kecepatan dan interpretabilitas, sementara Random Forest menunjukkan performa yang stabil di
berbagai skenario.
Kata kunci: ujaran kebencian, logistic regression, random forest, tf-idf, n-gram.
Evaluation of Logistic Regression and Random Forest Algorithms for Hate Speech

Identification
Adam Julizar

Hate speech on social media poses a serious threat to digital well-being, particularly in linguistically
diverse contexts such as Indonesia. This study aims to evaluate and compare the performance of two
machine learning algorithms, Logistic Regression and Random Forest, in classifying Indonesian-language
hate speech, as well as to examine the impact of using TF-IDF and a combination of TF-IDF with N-Gram
as feature extraction methods on classification performance. The research utilizes an Indonesian-language
dataset from 2019, consisting of 11,122 annotated entries obtained from the Kaggle platform. The data
underwent several stages of analysis, including preprocessing (text cleaning, normalization, stopword
removal, and stemming), feature extraction using TF-IDF and TF-IDF + N-Gram, and classification using
Logistic Regression and Random Forest algorithms. The results show that the combination of TF-IDF +
N-Gram improves classification performance, with Logistic Regression accuracy increasing from 81% to
83%, and Random Forest from 82% to 83%. The recall of Logistic Regression also increased from 79% to
85%. Logistic Regression excels in terms of speed and interpretability, while Random Forest demonstrates
stable performance across various scenarios.
Keywords: hate speech, logistic regression, random forest, tf-idf, n-gram.