A Proficient Heuristic-Based Routing Algorithm for Vehicular Mesh Networks Using Random Forest Link Stability Prediction

Dharanyadevi Pandian¹ , Palani Dhandapani², Devi Arumugam³ and Xavier Fernando⁴
1. Department of CSE, Puducherry Technological University, Puducherry, India
2. Department of ECE, University College of Engineering Villupuram, Villupuram, India
3. Department of ECE, IFET College of Engineering, Villupuram, India
4. Intelligent Computing and Communications Lab, Toronto Metropolitan University, Toronto, ON, Canada
Correspondence to: Dharanyadevi Pandian, dharanyadevi@gmail.com

DOI: https://doi.org/10.70389/PJS.100233

Additional information

• Ethical approval: N/a
• Consent: N/a
• Funding: No industry funding
• Conflicts of interest: N/a
• Author contribution: Dharanyadevi Pandian, Palani Dhandapani, Devi Arumugam and Xavier Fernando – Conceptualization, Writing – original draft, review and editing
• Guarantor: Dharanyadevi Pandian
• Provenance and peer-review: Unsolicited and externally peer-reviewed
• Data availability statement: N/a

Keywords: Vehicular mesh networks, Heuristic fitness-based routing, Random forest link stability prediction, Queue-aware relay selection, Relay suit-ability index.

Peer Review
Received: 6 November 2025
Last revised: 17 December 2025
Accepted: 17 December 2025
Version accepted: 4
Published: 16 January 2026

Introduction

The Vehicular Ad-Hoc Networks (VANETs) represents an evolving technology designed to support the rapid development of advanced vehicular applications. The increasing growth of novel applications and services within vehicular environments has gained significant interest from the research community.^1–7 VANETs are categorized by their widespread, infrastructure-less and lack of centralized control.^8,9 These networks facilitate radio communication between moving vehicles using Dedicated Short Range Communication (DSRC) technology.¹⁰ Wireless Mesh Networks (WMNs), are multi-hop communication networks with self-organizing nodes that operate without a centralized coordinator.¹¹ Vehicular Mesh Network (VMesh) is a hybrid networking paradigm that incorporates the features of both VANETs and WMNs. It enables dynamic networking among vehicles, where nodes (vehicles) can be added, updated, or removed.

However, the VMesh environment causes several obstacles due to its intrinsic properties, such as dynamic vehicle mobility, frequent topology changes, lack of centralized management, non-uniform node density and neighborhood configurations, use of broadcasting and geocasting communication, congested and unreliable wireless channels, hidden terminal issues, limited resource capacity, and risk of security attacks. The motivation of this article  are as follows: Modern intelligent transportation systems, which advance real-time transmission between moving vehicles, however the network topology changes, connectivity interruptions and packet loss are facing major issues in vehicular milieu. Existing methodologies often struggle due to rapid and unpredictable changes in the environment which leads to packet loss, congestion and longer delays. To overcome these issues this article estimates the reliability of links, adapt to changing network conditions and maintain reliable data transmission. The main contributions of the article are as follows:

1. This article introduces the proficient routing algorithm to optimize routing in the VMesh network which combines the heuristic fitness functions and random forest-based link prediction.
2. This article focuses on several routing parameters such as, link stability (predicted by random forest), distance, packet delivery ratio, signal strength, and dynamic valuation and selection of the most suitable intermediate vehicles (IVs).
3. Random Forest is incorporated exclusively to predict the stability of potential communication links, enabling the routing algorithm to select relay nodes that offer more reliable and long-lasting connections.
4. This method improves packet delivery, better uses available bandwidth, increases success rates, supports scalability, and ensures better network performance than previous routing methods.

The rest of this article is structured as follows. The related works and comparison between the existing and proposed work are discussed in Section ‘Related Works’ and Section ‘A Proficient Heuristic-Based Routing Algorithm for VMeshs Using Random Forest Link Stability Prediction’ describes the proposed Proficient Routing Algorithm using Heuristic Functions and Random Forest (PRA-HF-RF) for VMesh milieu. Section ‘Simulation and Results’ gives the simulation results and analysis of experimental results based on the metrics. Section ‘Conclusion’ concludes this article.

Related Works

A few efficient routing process have been proposed to tackle challenges such as gridlock, latency, packet loss, and protocol overhead in the VMesh environment. Gupta et al.¹² discuss the routing issues in WMNs as an optimization issue for both stable and active network conditions. According to the authors although numerous adaptive routing process have been developed, they normally lack a solid theoretical basis for evaluating overall network proficiency. Tang et al.¹³ introduces a bandwidth allocation method which intended at enhancing routing performance. However, Gupta et al.¹² argue that while these methods focus on achieving proficient bandwidth usage, they often supervise the traffic load on transmitting nodes and, as a result, fail to account for the actual demand on network resources. To address this gap, Gupta introduced a strategy that treats the routing challenge as an optimization problem under both static and dynamic conditions.

Their main goal was to improve the balance between data flow and its required resources, while also ensuring that the network remains fair and can manage scheduling effectively. Mamatha et al.¹⁴ employed the SHA-1 algorithm to accomplish information exchange between road side unit (RSU) and vehicles and the support vector machine (SVM) algorithm for road condition analysis and pothole detection. The authors suggested in¹⁵ an ensemble stacking-based machine learning (ML) model with a booster model and combining several ML models. In order to increase network traffics precision and effectiveness. Although dynamic VANETs have advanced significantly as a result of these studies significant computational complexity was also added especially in ML-based approaches. This complexity may restrict these techniques real-time adaptability and scalability in environments with limited resources and rapid change which could impede their practical implementation and the full realization of their theoretical advantages in VANETs.

Prior studies have explored various congestion-control and routing-optimization techniques for vehicular networks. Some ML driven approaches improve accuracy but introduce significant computational overhead, making real-time deployment difficult. Routing schemes based solely on mobility or signal measurements frequently overlook queue stability and load distribution. Furthermore, traditional congestion mitigation mechanisms generally react after performance degradation occurs. Unlike those methods, PRA-HF-RF prevents congestion proactively by combining predictive analytics and heuristic evaluation, ensuring a more consistent forwarding path.

A Proficient Heuristic-Based Routing Algorithm for VMeshs Using Random Forest Link Stability Prediction

The proposed algorithm is designed to proficiently identify the most suitable relay node and Base Station (BS) in a VMesh network. As defined in Equation 1, the Proficient Routing Algorithm Heuristic Fitness (PRA-HF) assesses each proficient relay node based on three key parameters:

• Every vehicle has a (queue buffer) to temporarily stores the packets it needs to forward. If the queue is full, new packets will be dropped or delayed. If the queue has more free space, the node can handle more incoming data without congestion. This research takes into account the free space in the transmission queue of nodes to enhance data forwarding efficiency.
• This study incorporates retrospective packet delivery ratio (PDR) values to ensure that the selected relay nodes contribute to maintaining a high rate of successful packet delivery. This research uses the acknowledgment (ACK)-based feedback, the node receives acknowledgments from the destination or the next hop and stores the delivery success rate in a local buffer. The value is periodically updated with exponential weighting to capture recent performance changes.
• To improve routing efficiency, the shortest distance from the current node to the destination is employed as a selection criterion.

The PRA-HF function is expressed as:

where,

• ‘Q_f (i)’ is the available space in the transmission queue of the relay node.
• ‘PDR(i)’ is the retrospective packet delivery ratio of node.
• ‘D_sd(i)’is the Euclidean distance between the source and the destination.
• ‘α, β, γ’ is Heuristic weights; α+ β + γ = 1. α = 0.3, β = 0.4, γ = 0.3, weight are selected through the grid search.
• α, β, γ € [0,1]

Queue Management Mechanism

Each vehicle in the VMesh network has:

• An input buffer queue to store incoming packets.
• An output buffer queue to hold packets ready for transmission.

The First In, First Out (FIFO) queuing process in the VMesh network is as follows:

• Packets are handled in the order of arrival.
• If the queue overflows, packets are dropped from the tail, ensuring fair resource usage.

This process supports proficient traffic handling, while PRA-HF confirms intelligent IV selection under changing traffic conditions. The data transmission in the VMesh milieu is limited to the vehicle concentration zone, which defines the spatial range within which nearby vehicles are able to send and receive the requests and responses. The Equation 2 determines the convergence area are as follows:

The Equation 3 estimates the number of vehicles within the convergence:

where,

• ‘r’ represents the proficient communication radius of a vehicle
• ‘V_d’ is the Vehicle density (vehicles/km²)
• ‘A_Con’ is the estimated convergence area with active communication, which typically ranges between 100 and 300 m.
• ‘N_Con’ is the expected number of vehicles within the communication zone.

The convergence metric help to governs the set of vehicles that are eligible to participate in routing decisions based on its communication range and proximity. The three primary routing cases are as follows:

Case 1: Routing Based on BS Availability

As illustrated in Figure 1 and Equation 4, if a BS is available within the vehicle’s convergence range, the source vehicle directly transmits the data packet to the BS. This allows for faster and more reliable communication without involving IVs.

In contrast, as shown in Figure 2 and Equation 4, if no BS is available within the convergence range, the source vehicle forwards the packet through IVs using multi-hop communication.

This scenario is represented as:

where,

• ‘Dest’ is Destination node (BS or IV)
• ‘BS’ is Base Station
• ‘IV’ is Intermediate Vehicle,
• ‘A_Con’ is Vehicle’s convergence area.

This approach helps the algorithm select the shortest and most suitable path to the destination.

Case 2: Routing Via IVs

If the BS cannot be reached directly, the data packet is forwarded through IV. As shown in Equation 5, the routing is influenced by the number of IV present within the convergence area.

• If only one IV is available, the packet is forwarded directly to that node.
• If there is more than one IVs, the optimal IVs is selected based on the PRA-HF value with respect to the metrics such as, distance, retrospective packet delivery ratio and queue space.

This routing case is expressed as:

The selected node serves as an intermediate relay, and this process repeats until the data reaches the BS. After the BS receives the request, it forwards it to the service provider. The response from the service provider returns through the same path in order ensure reliability.

Case 3: Routing Based on Link Stability Using Random Forest

As given in Equation 6, when the difference in PRA-HF values among neighbor nodes is less than 10%, it shows a high vehicle density scenario, where multiple vehicles are equally eligible for transmission. In this scenario, the traffic load is high, and supplementary measures such as link stability become vital to guarantee reliable and proficient data transmission. This condition can be expressed as:

where,

• PRA-HF(i) and PRA-HF(j) are the fitness values of two neighbor nodes.
• The absolute relative difference is less than 10%.

Therefore, a Random Forest-based Link Stability Prediction Model is employed to distinguish the most stable link among the comparable neighbors. The proposed model evaluates the link stability score (LS) using the parameters as follows:

• Received Signal Strength Indicator (RSSI),
• Vehicle speed and relative velocity,
• Mobility pattern,
• Retrospective link duration.

These features are fed into the Random Forest model, which has been trained on labeled data (e.g., past links labeled as stable or unstable). The model learns patterns and assigns a predicted LS in the range of 0–1, where higher values indicate stronger and more reliable links. The Random Forest algorithm builds an ensemble of multiple decision trees, each trained on a random subset of the data and features (bagging). During training, each tree learns a set of decision rules to correlate the input parameters with link stability outcomes. Random Forest outputs a stability probability between 0 and 1. The decision metric for IV selection is designed to combine both the PRA-HF heuristic fitness value and the predicted link LS. The final score, referred to as the Relay Suitability Index (RSI), is computed using a weighted combination as given in Equation 7:

Where:

• ‘RSI(i)’ is Relay Suitability Index is used to select the optimal relay node,
• ‘PRA-HF’ is the heuristic fitness value based on queue space, PDR and distance.
• ‘LS(i)’ is the Link Stability (Random Forest output).
• ‘μ and ρ’ are weighting coefficients that satisfy μ +ρ = 1. μ = 0.6, ρ = 0.4 via grid search.

The node with the RSI is selected. This ensures reliable routing, especially in dynamic scenarios. The routing condition for Case 3 is:

As expressed in Equation 8, if there are multiple IVs among the neighbors whose PRA-HF values differ by less than 10%, then RSI is used to select the best node. Otherwise, the packet is forwarded to the node with the highest PRA-HF value. In case by integrating PRA-HF-RF the routing mechanism achieves:

• Lower packet loss
• Higher throughput
• Improved reliability in rapidly changing vehicular networks.

The proposed process which ensures stability, reliability, intelligence and high-performance in the VMesh milieu is explained in Algorithm 1.

Simulation and Results

Deploying and evaluating a VMesh network in real traffic conditions is difficult, expensive, and time-consuming. To overcome these limitations, simulation environments are used because they provide a controlled, repeatable, and cost-efficient platform for assessing routing performance before real-world deployment. In this work, the proposed PRA-HF-RF routing scheme was implemented and tested using the Traffic and Network Simulation Environment (TraNS) framework, which links mobility, mapping, and network simulation components. The experimental setup are as follows:

• Mobility Framework: SUMO 1.19 using the TraCI interface.
• Map Source: A 2 km × 2 km urban area taken from OpenStreetMap (OSM).
• Network Simulator: NS-3.40.
• ML Pipeline: Python 3.11 and scikit-learn 1.3, with offline Random Forest stability results supplied to NS-3 via lookup tables.

The Random Forest model is trained offline using mobility traces generated from the SUMO-NS-3 milieu. Each link sample has a features such as RSSI, relative speed, inter-vehicle distance, node density, queue availability, packet delivery ratio, and prior link duration. Each link instance is classified as stable or unstable according to whether it remained active past the specified stability threshold. After training, the model produces a set of stability values that NS-3 loads at the start of the simulation. During routing, each node checks the real-time details of its neighbors and finds the LS that matches them (between 0 and 1). This score is then used along with the PRA-HF metric to pick the most reliable relay vehicle. The training data contains several thousand link samples collected under different vehicle speeds, traffic levels, and communication conditions. The data is split into training, validation, and testing sets in a 70/15/15 ratio. The important parameters such as number of trees, the depth of each tree, and the number of input fields are adjusted through k-fold cross-checks to improve prediction accuracy. All routing protocols are tested under the same simulation conditions to allow a fair comparison and to confirm the performance gains of the proposed PRA-HF-RF.

Evaluation Metrics

The performance of PRA-HF-RF is evaluated using the metrics are described as follows:

• Packet Collision Ratio (PCR): It indicates network congestion and interference.
• Signaling Load: It indicates the overhead caused by control messages during the routing process.
• Success Rate: It indicates the number of packets successfully delivered to the destination.

Algorithm 1 | PRA-HF-RF

Input: S: Source vehicle node Destination: Base Station ACon: Vehicle’s convergence range Queue_Info: Queue buffer status (input/output) Node_List: List of neighboring nodes within ACon α, β, γ: Weight factors for PRA-HF µ and ρ Weight coefficients RF_Model: Pre-trained Random Forest model Begin if BS ∈ ACon then // Case 1: Direct Base Station Availability Send Packet(Source, Destination) return [Source, Destination] else //Identify Intermediate Vehicles (IV) within ACon IV_List ← Get Intermediate Vehicles(Source, ACon) Calculate NCon = Vd*ACon if Length(IV_List) == 1 then // Case 2: Single Intermediate Vehicle Relay ← IV_List[0] else if Length(IV_List) > 1 then // Case 3: Multiple Intermediate Vehicles for each node in IV_List do PRA ← Compute PRA-HF(i) = αQf(i) + βPDR(i) + γ(1 / Dsd(i)) // Check for comparable PRA-HF values Comparable ← False For all unique pairs (i, j) in IV_List: If abs(PRA-HF[i] – PRA-HF[j]) / PRA-HF[i] ≤ 0.1: Comparable ←True Break If Comparable == True: // Case 3: PRA-HF values are comparable, use RSI For each node i in IV_List: Extract features the RSSI[i], Speed[i], RelVelocity[i], Mobility[i], LinkDuration[i] LS[i] = Predict_Stability_Score(RandomForestModel, features[i]) RSI(i) = μ * PRA-HF(i) + ρ * LS(i) SelectedNode = node with max(RSI[i]) Else: // PRA-HF values are not comparable, use max PRA-HF SelectedNode = node with max(PRA-HF[i]) Forward packet to SelectedNode Function Predict_Stability_Score(model, features): // Use trained Random Forest model to predict link stability score (LS) Return model.predict(features) End

PCR

The distance between the adjacent Internet Gateways (IGWs) is 3.6 km. As shown in Figure 3, the PCR for both bandwidth utilization and fairness enhancement – medium access control (BUFE-MAC) and the proposed PRA-HF-RF increases with vehicle density. However, the proposed PRA-HF-RF shows better performance by proficiently managing collisions by its heuristic fitness function and queue-based scheduling. Furthermore, the integration of Random Forest-based link stability prediction further improves route consistency, and reduces redundant retransmissions. As a result, the PCR of PRA-HF-RF remains lower than the BUFE-MAC.

Signaling Load

As shown in Figure 4, the signaling load for both BUFE-MAC and the proposed PRA-HF-RF increases with the rise in vehicle velocity. This is due to the frequent topology changes that occur at higher speeds, requiring more control messages for route maintenance. However, the proposed PRA-HF-RF keeps a lower signaling load compared to BUFE-MAC by assuring that packets are transmitted within the optimal coverage area and also by transmitting to the stable link. This guarantees the PRA-HF-RF more proficient in handling the dynamic environment.

Success Rate

As illustrated in Figure 5, the success rate of proposed and existing decreases as the number of requests increases. This decreases is due to packet loss, delay, increasing congestion, and protocol overhead caused by the increased density. However, the proposed PRA-HF-RF attains a higher success rate than BUFE-MAC by proficiently managing packet transmission using a heuristic fitness function that selects optimal routing path. Furthermore, the amalgamation of Random Forest-based link stability prediction improves route reliability by selecting stable links, thereby minimalizing interruptions and guaranteeing a higher number of successful communications. As a outcome, PRA-HF-RF outperforms BUFE-MAC.

Conclusion

This article shows that PRA-HF-RF offers a reliable and highly proficient routing mechanism for dynamic and high-density VMesh environments. By integrating a heuristic function with link stability prediction, the proposed PRA-HF-RF consistently identifies reliable relay paths, lessens redundant retransmissions, and reduces route interruptions. This article proves the strength of the PRA-HF-RF routing mechanism through the performance metrics.

References

1. Abbas AH, Ahmed AJ, Rashid SA, Abdulsattar NF, Hassan MH, Jubair MA, et al. Predictive hybrid routing with multi objective optimization model for opportunistic routing for VANET. In: 2022 5th international conference on engineering technology and its applications (IICETA); 2022. p. 470–6. http://doi.org/10.1109/IICETA54559.2022.9888274
2. Joshi J, Singh AK. Performance analysis of routing protocols for vehicular networks. Int J Veh Inf Commun Syst. 2024;9(1):60–80. https://doi.org/10.1504/IJVICS.2024.136276
3. Hung LL, Chang CY, Chen CC, Chen YC. BUFE-MAC: a protocol with bandwidth utilization and fairness enhancements for mesh-backbone-based VANETs. IEEE Trans Veh Technol. 2012;61(5):2208–21. http://doi.org/10.1109/TVT.2012.2189592
4. Taha MMI, Hasan YMY. VANET-DSRC protocol for reliable broadcasting of life safety messages. In: Signal processing and information technology, 2007 IEEE international symposium on; 2007. p. 104–9. http://doi.org/10.1109/ISSPIT.2007.4458046
5. Kaja H, Stoehr JM, Beard C. V2X-assisted emergency vehicle transit in VANETs. Simulation. 2024;100(3):229–44. https://doi.org/10.1177/00375497231209774
6. Taleb T, Sakhaee E, Jamalipour A, Hashimoto K, Kato N, Nemoto Y. A stable routing protocol to support ITS service in VANET networks. IEEE Trans Veh Technol. 2007;56(6):3337–47. http://doi.org/10.1109/TVT.2007.906873
7. Sataraddi MJ, Kakkasageri MS. Machine learning based vehicle-to-infrastructure communication in VANETs. in proceedings of the 2021 IEEE 18th India Council International Conference (INDICON 2021); 2021. p. 1–6. http://doi.org/10.1109/INDICON52576.2021.9691710
8. Bernsen J, Manivannan D. Unicast routing protocols for vehicular ad hoc networks: a critical comparison and classification. Pervasive Mobile Comput. 2009;5(1):1–18.https://doi.org/10.1016/j.pmcj.2008.09.001
9. Ghosh J, Kumar N, Al-Utaibi KA, Sait SM, Vo VN, So-In C. Reliable data transmission for a VANET-IoIT architecture: a DNN approach. Int Things (Netherlands). 2024;25:101129. https://doi.org/10.1016/j.iot.2024.101129
10. Al-Sultan S, Al-Doori MM, Al-Bayatti AH, Zedan H. A comprehensive survey on vehicular ad hoc networks. J Netw Comput Appl. 2014;37:380–92. https://doi.org/10.1016/j.jnca.2013.02.036
11. Alotaibi E, Mukherjee B. A survey on routing algorithms for wireless ad-hoc and mesh networks. Comput Net. 2012;56:940–65. http://doi.org/10.1016/j.comnet.2011.10.011
12. Gupta BK, Patnaik S, Mallick MK, Nayak AK. Dynamic routing algorithm in wireless mesh network. Int J Grid Utility Comput. 2017;8(1):53–60. https://doi.org/10.1504/IJGUC.2017.082144
13. Tang J, Xue G, Zhang W. Maximum throughput and fair bandwidth allocation in multi-channel wireless mesh networks. In: Proceedings IEEE INFOCOM 2006. 25th IEEE international conference on computer communications; 2006. p. 1–10. http://doi.org/10.1109/INFOCOM.2006.249
14. Mamatha G, Sharan HS, Prathik R, Priya DS, Prajwal U. Smart vehicular communication for road status analysis and vehicle trajectory prediction. In: 2020 third international conference on smart systems and inventive technology (ICSSIT); 2020. p. 1081–7. http://doi.org/10.1109/ICSSIT48917.2020.9214252
15. Amiri PAD, Pierre S. An ensemble-based machine learning model for forecasting network traffic in VANET. IEEE Access. 2023;11:22855–70. http://doi.org/10.1109/ACCESS.2023.3253625

Cite this article as:
Dharanyadevi P, Palani D, Devi A and Fernando X. A Proficient Heuristic-Based Routing Algorithm for Vehicular Mesh Networks Using Random Forest Link Stability Prediction. Premier Journal of Science 2026;16:100233

Export to EndNote Export to Mendeley