Klasifikasi Berita Pada Laman Website TurnBackHoax.id Menggunakan Algoritma Random Forest

Berita hoax merupakan hasil akhir dari berita yang telah direkayasa kebenarannya. Tersebarnya hoax dimasyarakat akan menimbulkan pertengkaran, perdebatan, serta perpecahan antar pihak, baik secara individu maupun kelompok tertentu. Tersebarnya berita hoax ini perlu dilakukan antisipasi secepat mungkin. Pemerintah bersama Masyarakat Anti Fitnah dan Hoax Indonesia (MAFINDO) telah melakukan antisipasi dan pencegahan tersebarnya berita hoax dengan menciptakan laman TurnBackHoax.id yang bisa di akses oleh siapapun dan dimanapun secara online. Pengklasifikasian berita hoax yang dilakukan oleh MAFINDO masih dilakukan secara manual, sehingga membutuhkan waktu dan tenaga lebih banyak. Penelitian yang dilakukan bertujuan untuk mengklasifikasikan berita yang berkembang dimasyarakat kedalam kategori hoax atau fakta. Penelitian ini akan menunjukkan hasil ketepatan klasifikasi berita yang masuk kategori hoax atau fakta, sehingga dapat diketahui performa dari model dan hasil prediksi yang telah terbentuk. Data yang digunakan pada penelitian ini merupakan data sekunder dari hasil scrapping berita pada website TurnBackHoax.id dengan mengambil judul berita, waktu terbit berita, dan sekilas isi berita. Metode klasifikasi yang digunakan pada penelitian ini yaitu algoritma Random Forest. Hasil analisis menyimpulkan bahwa evaluasi performa dari data training diperoleh nilai akurasi sebesar 86,8%, sensitivity sebesar 97,6%, specificity sebesar 67,6%, dan nilai AUC sebesar 0,826 sedangkan untuk data testing diperoleh nilai akurasi sebesar 86,8%, sensitivity sebesar 97,7%, specificity sebesar 67,4%, dan nilai AUC sebesar 0,826. Performa data training dan testing menunjukkan model terbukti mampu untuk mengklasifikasi berita hoax atau fakta karena memiliki nilai akurasi dengan kategori good excellent dan didukung nilai AUC yang masuk kategori baik, serta nilai sensitivity yang lebih besar daripada nilai specificity. Hasil performa dari data training dan data testing tidak memiliki error yang tinggi, sehingga tidak terjadi underfitting atau overfitting ================================================================================================================================== Hoax news is the end result of news that has been engineered to be true. The spread of hoaxes in society will lead to quarrels, debates, and divisions between parties, both individually and in certain groups. The spread of hoax news needs to be anticipated as soon as possible. The government together with the Indonesian Anti Defamation and Hoax Society (MAFINDO) have anticipated and prevented the spread of hoax news by creating the TurnBackHoax.id page which can be accessed by anyone and anywhere online. The classification of hoax news by MAFINDO is still done manually, so it requires more time and effort. The research conducted aims to classify news that develops in the community into the category of hoaxes or facts. This research will show the results of the accuracy of the classification of news that fall into the category of hoaxes or facts, so that the performance of the model and prediction results that have been formed can be known. The data used in this study is secondary data from the results of scrapping data on the TurnBackHoax.id website by taking the news title, news publication time, and news content at a glance. The classification method used in this study is the Random Forest algorithm. The results of the analysis concluded that the performance evaluation of the training data obtained an accuracy value of 86.8%, a sensitivity of 97.6%, a specificity of 67.6%, and an AUC value of 0.826 while for data testing an accuracy value of 86.8% was obtained. sensitivity of 97.7%, specificity of 67.4%, and AUC value of 0.826. The performance of the training and testing data shows that the model is proven capable of classification hoax news or facts because it has an accuracy value in the good excellent category and is supported by the AUC value which is in the good category, and the sensitivity value is greater than the specificity value. The performance results from the training data and data testing do not have high errors, so there is no underfitting or overfittin