Prediksi RSRP 4G LTE Kota Padang Menggunakan Random Forest Regresion Berbasis Data Drive Test

Abstract

Abstract

The development of 4G LTE networks in Padang City necessitates accurate evaluation of signal coverage quality to support network optimization and quality of service management. Reference Signal Received Power (RSRP) is the primary parameter for assessing signal strength received by User Equipment (UE) from eNodeB. RSRP values obtained through drive testing are highly fluctuating due to radio propagation conditions, urban obstacles, and inter-cell interference, making conventional approaches insufficient for modeling the nonlinear signal distribution. This study applies a Random Forest Regression model to predict RSRP based on drive test data collected across Padang City, a region with diverse topography ranging from coastal areas to highland terrain. A dataset of 32,454 data points was obtained after preprocessing from 33,682 raw records. Input variables consisted of latitude, longitude, RSRQ, and SINR, with RSRP as the target variable. The dataset was divided into 70% training and 30% testing subsets (random_state = 42). Model performance was evaluated using MAE, RMSE, MSE, and R². Results yielded MAE of 3.46 dBm, RMSE of 4.71 dBm, and R² of 77.43%, indicating the model reliably captures the majority of actual RSRP variation. Spatial coverage mapping revealed Good-to-Very Good signal quality dominating the city center, while Fair-to-Bad categories concentrated in peripheral areas such as Lubuk Kilangan and Lubuk Minturun. These findings confirm Random Forest Regression as a viable data-driven approach for RSRP prediction and spatial network coverage analysis in geographically diverse urban regions.

Abstrak

Perkembangan jaringan 4G LTE di Kota Padang mendorong kebutuhan evaluasi kualitas cakupan sinyal secara akurat guna mendukung optimasi jaringan dan manajemen kualitas layanan. RSRP merupakan parameter utama untuk mengukur kekuatan sinyal yang diterima UE dari eNodeB. Nilai RSRP hasil drive test bersifat fluktuatif akibat kondisi propagasi radio, hambatan lingkungan perkotaan, dan interferensi antar sel, sehingga pendekatan konvensional tidak memadai untuk memodelkan distribusi sinyal nonlinier. Penelitian ini menerapkan model Random Forest Regression untuk memprediksi nilai RSRP berbasis data drive test di Kota Padang, wilayah dengan topografi beragam mulai dari pesisir pantai hingga kawasan perbukitan. Dataset 32.454 titik data diperoleh setelah pra-pengolahan dari 33.682 data mentah. Variabel input: latitude, longitude, RSRQ, dan SINR; variabel target: RSRP. Dataset dibagi 70% latih dan 30% uji (random_state = 42). Kinerja model dievaluasi menggunakan MAE, RMSE, MSE, dan R². Hasil: MAE 3,46 dBm, RMSE 4,71 dBm, R² 77,43%, mengindikasikan model merepresentasikan sebagian besar variasi RSRP aktual. Pemetaan spasial menunjukkan dominasi Good hingga Very Good di pusat kota, sementara Fair hingga Bad di wilayah pinggiran seperti Lubuk Kilangan dan Lubuk Minturun. Temuan ini mengonfirmasi Random Forest Regression sebagai pendekatan berbasis data yang andal untuk prediksi RSRP dan analisis cakupan spasial pada wilayah dengan kondisi geografis beragam.