Electronic Theses and Dissertation

Peningkatan data yang dihasilkan pada era big data telah menghadirkan tantangan baru sekaligus peluang dalam pengembangan dan penerapan teknologi machine learning. Penelitian ini bertujuan untuk membandingkan performa tiga metode feature selection, yaitu filter method, wrappers method, dan embedded method pada model random forest untuk mengklasifikasikan rumah tangga rawan pangan di Provinsi Aceh tahun. Hasil penelitian menunjukkan bahwa model klasifikasi random forest dengan embedded method pada rasio data 90:10 memiliki performa tertinggi. Model ini menghasilkan nilai accuracy sebesar 78,35%, nilai precision sebesar 81,46%, recall sebesar 96,64%, dan f1-score sebesar 87,55%. Kemudian, hasil feature importance dari model terbaik mengidentifikasikan lima variabel dengan tingkat kepentingan tertinggi dalam menentukan status rawan pangan, yaitu pendidikan Kepala Rumah Tangga (KRT), sumber air minum, jumlah rumah tangga usia 0-14 tahun. Kebiasaan merokok, dan status bangunan tempat tinggal. Hasil penelitian menunjukkan bahwa embedded method merupakan metode feature selection terbaik yang dapat meningkatkan nilai accuracy model klasifakasi random forest dan memberikan wawasan penting terkait variabel-variabel yang berkontribusi terhadap status rumah tangga rawan pangan di Provinsi Aceh tahun 2022.

Kata kunci: Feature Selection, Filter Method, Wrappers Method, Embedded Method, Klasifakasi Random Forest, Rawan Pangan, Aceh.

The increase in data generated in the big data era has presented new challenges as well as opportunities in the development and application of machine learning technology. This research aims to compare the performance of three feature selection methods: filter method, wrappers method, and embedded method on random forest models to classify food-insecure households in Aceh Province. The results showed that the random forest classification model with embedded method at a 90:10 data ratio had the highest performance. Model produced an accuracy value of 78.35%, precision value of 81.46%, recall of 96.64%, and f1-score of 87.55%. Then, the feature importance results from the best model identified five variables with the highest importance in determining food insecurity status, namely the education of the Head of Household (KRT), drinking water source, number of household members aged 0-14 years, smoking habits, and residential building status. The research results indicate that the embedded method is the best feature selection method that can improve the accuracy value of the random forest classification model and provide important insights regarding variables that contribute to household food insecurity status in Aceh Province in 2022. Keywords: Feature Selection, Filter Method, Wrappers Method, Embedded Method, Random Forest Classification, Food Insecurity, Aceh