Electronic Theses and Dissertation

Perkembangan teknologi yang pesat telah mengubah industri otomotif, terutama dalam
meningkatkan ketersediaan dan permintaan mobil bekas. Platform daring seperti OLX
Autos menjadi semakin populer, terutama selama pandemi, karena memudahkan
transaksi antara pembeli dan penjual. Namun, menetapkan harga mobil bekas tetap
menjadi tantangan. Oleh karena itu, machine learning dapat menjadi salah satu solusi
yang dapat dilakukan, khususnya algoritma Extreme Gradient Boosting (XGBoost) dan
Random Forest, untuk memprediksi harga mobil bekas. XGBoost mengurangi
overfitting dan meningkatkan akurasi prediksi melalui teknik regularisasi, sementara
Random Forest menggabungkan beberapa model untuk menghasilkan prediksi yang
lebih akurat dan stabil. Dataset yang digunakan merupakan data hasil crawling dan
scraping situs mobil bekas OLX Indonesia. Hasil evaluasi prediksi menunjukkan
bahwa model Random Forest memiliki performa yang sangat baik dengan konfigurasi
hyperparameter n_estimators = 100, random_state = 0, min_samples_leaf = 1,
max_features = 0.3, n_jobs = -1, dan oob_score = True mencapai nilai R-squared
sebesar 0.9485 dengan waktu pengujian sebesar 120 ms.

Rapid technological advancements have significantly transformed the automotive industry, particularly by increasing the availability and demand for used cars. Online platforms such as OLX Autos have become increasingly popular, especially during the pandemic, as they facilitate transactions between buyers and sellers. However, setting the price of used cars remains a challenge. Therefore, machine learning can be a viable solution, specifically using algorithms like Extreme Gradient Boosting (XGBoost) and Random Forest, to predict the prices of used cars. XGBoost reduces overfitting and improves prediction accuracy through regularization techniques, while Random Forest combines multiple models to produce more accurate and stable predictions. The dataset used comprises data obtained from crawling and scraping the OLX Indonesia used car website. The prediction evaluation results indicate that the Random Forest model shows excellent performance with the hyperparameter configuration of n_estimators = 100, random_state = 0, min_samples_leaf = 1, max_features = 0.3, n_jobs = -1, dan oob_score = True, achieving an R-squared value of 0.9485 with a testing time of 120 ms.