A Bayesian-Optimised Heterogeneous Stacked Ensemble Framework for Accurate Path Loss Prediction in Millimetre-Wave Wireless Networks

Purpose: This paper proposes a hybrid machine learning model to achieve accurate path loss estimation to significantly improve network planning, spectrum utilisation, and energy efficiency in next-generation wireless systems. Design/Methodology/Approach: A hybrid model integrating Random Forest (RF), Support Vector Regression (SVR), and Deep Neural Network (DNN) is developed alongside baseline models (Gradient Boost Regression, Multi-layer Perceptron, and XGBoost). The dataset is generated using the 3GPP channel model. The 3GPP TR 38.901 UMi and UMa setting is considered with a frequency of 28 GHz, Tx-Rx distance of 10-300 m, and path loss in dB predicted. There are 5000 data points obtained through simulation of 3GPP model. Since the labels are simulator-generated, the high may be optimistic. Preprocessing includes outlier removal via Isolation Forest, RSSI smoothing using a Kalman filter, and spatial cross-validation for data splitting. Hyperparameter optimisation is executed using a Bayesian optimiser. Model performance is evaluated using MAE, score, and RMSE. The 95th and 99th percentile absolute errors, error variation with distance, and performance will be evaluated individually for LOS and NLOS cases. Research Limitation: The evaluation relies on synthetic channel data generated via the 3GPP standard model; empirical validation through real-world mmWave field measurements across varied geographical topologies remains an area for extended verification. Findings: The proposed hybrid model outperforms comparative models across all evaluated frequencies and scenarios. It achieves a training performance of , , and , alongside a validation performance of , , and . Practical Implication: Accurate path loss and LOS probability estimation lowers infrastructure deployment expenditure, optimises base station placement, conserves transmit power, and enhances spectrum management for cellular operators rolling out dense mmWave networks. Social Implication: Reliable mmWave propagation modelling supports robust connectivity, lowers network energy footprints, and provides high Quality of Service (QoS) essential for mission-critical Internet of Things (IoT), dense urban mobile networks, and smart city infrastructure. Originality/Value: The work presents a novel hybridisation of RF, SVR, and DNN optimised via Bayesian search, combined with spatial cross-validation and Kalman filtering, outperforming individual state-of-the-art regression models across diverse mmWave operational environments.

Authors

Institutions

Publication Details

Journal
African Journal Of Applied Research
Published
2026-11-18
DOI
https://doi.org/10.26437/zk1g1509
Primary Topic
Millimeter-Wave Propagation and Modeling
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

A Bayesian-Optimised Heterogeneous Stacked Ensemble Framework for Accurate Path Loss Prediction in Millimetre-Wave Wireless Networks

A. Madankar, A. Khobragade
1 links
African Journal Of Applied Research
Millimeter-Wave Propagation and Modeling
article

A Bayesian-Optimised Heterogeneous Stacked Ensemble Framework for Accurate Path Loss Prediction in Millimetre-Wave Wireless Networks

A. Madankar, A. Khobragade
article en

Abstract

Purpose: This paper proposes a hybrid machine learning model to achieve accurate path loss estimation to significantly improve network planning, spectrum utilisation, and energy efficiency in next-generation wireless systems. Design/Methodology/Approach: A hybrid model integrating Random Forest (RF), Support Vector Regression (SVR), and Deep Neural Network (DNN) is developed alongside baseline models (Gradient Boost Regression, Multi-layer Perceptron, and XGBoost). The dataset is generated using the 3GPP channel model. The 3GPP TR 38.901 UMi and UMa setting is considered with a frequency of 28 GHz, Tx-Rx distance of 10-300 m, and path loss in dB predicted. There are 5000 data points obtained through simulation of 3GPP model. Since the labels are simulator-generated, the high may be optimistic. Preprocessing includes outlier removal via Isolation Forest, RSSI smoothing using a Kalman filter, and spatial cross-validation for data splitting. Hyperparameter optimisation is executed using a Bayesian optimiser. Model performance is evaluated using MAE, score, and RMSE. The 95th and 99th percentile absolute errors, error variation with distance, and performance will be evaluated individually for LOS and NLOS cases. Research Limitation: The evaluation relies on synthetic channel data generated via the 3GPP standard model; empirical validation through real-world mmWave field measurements across varied geographical topologies remains an area for extended verification. Findings: The proposed hybrid model outperforms comparative models across all evaluated frequencies and scenarios. It achieves a training performance of , , and , alongside a validation performance of , , and . Practical Implication: Accurate path loss and LOS probability estimation lowers infrastructure deployment expenditure, optimises base station placement, conserves transmit power, and enhances spectrum management for cellular operators rolling out dense mmWave networks. Social Implication: Reliable mmWave propagation modelling supports robust connectivity, lowers network energy footprints, and provides high Quality of Service (QoS) essential for mission-critical Internet of Things (IoT), dense urban mobile networks, and smart city infrastructure. Originality/Value: The work presents a novel hybridisation of RF, SVR, and DNN optimised via Bayesian search, combined with spatial cross-validation and Kalman filtering, outperforming individual state-of-the-art regression models across diverse mmWave operational environments.

African Journal Of Applied ResearchVol. 12(8)
Yeshwantrao Chavan College of Engineering (IN)
Industry, innovation and infrastructure
Millimeter-Wave Propagation and Modeling
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon