Multivariate weighted isotonic regressive modest adaptive boosting-based resource-aware routing in WSN

WSN includes a scenario where many sensor nodes are distributed to monitor environmental conditions with route collected data toward sinks via the internet. WSNs efficiently manage the broader network with available resources, such as residual energy and wireless channel bandwidth. Therefore, a routing algorithm is essential to enhance battery-constrained networks. Many existing techniques are developed for balancing energy consumption, but the efficient routing was not achieved. The multivariate weighted isotonic regressive modest adaptive boosting-based Resource-Aware Routing (MWIRMAB-RAR) technique is introduced to enhance routing. The MWIRMAB-RAR technique includes a different process, namely resource-aware node selection, route path discovery, and data transmission. Initially, the MWIRMAB-RAR technique uses the modest adaptive boosting technique uses the multivariate weighted isotonic regression function for detecting resource-efficient sensor nodes for effective data transmission. After that, multiple route paths are established based on the time of flight method. Once after showing a route path, the source node sends data packets to the sink node via resource-efficient nodes. The data delivery was enhanced and minimized packet loss as well as delay. The simulation analysis is carried out on certain performance factors such as energy consumption, packet delivery ratio, packet loss rate, and delay with several data packets and sensor nodes. The obtained evaluation indicates that MWIRMAB-RAR outperforms well in increasing data packet delivery and reducing energy consumption, packet loss rate, and delay.


Introduction
WSNs are fundamental for various significant utilizations where a huge amount of sensor devices were developed in the monitoring field. Such kinds of sensor nodes suffer from restricted processing ability and energy. However, the nodes exist for extended time for enhancing the network lifetime during the data transmission to sink node via multi-hop routing. The sink node is dependable to make suitable decisions depending on received data. Energy conservation is a demanding problem in WSN. Several existing routing techniques were introduced, but balancing energy consumption and enhanced battery life of sensor nodes are still challenging issues in the distributed wireless network.
E3AF-based reliable routing algorithms were introduced in El-Fouly and Ramadan (2020) to improve data transmission consistency. However, it was not efficient to find better resource-efficient sensor nodes to further increase the network lifetime. Energy-efficient cooperative routing method for heterogeneous WSN (EERH) was developed in Hung et al. (2020) to extend the network lifetime. However, the designed routing failed to increase the reliable data delivery with minimum delay.
A routing strategy with a greedy algorithm was introduced in Puneetha and Kulkarni (2020) to attain a better network lifetime. However, the designed algorithm failed to apply the machine learning technique for improving the routing performance. DS-EERA was introduced in Tang et al. (2020) to enhance the data communication with minimum packet loss. However, the designed routing algorithm was inefficient in decreasing the end-to-end delay of data communication.
Energy-efficient routing-aware network coding-based data transmission was introduced in Singha and Nagaraju (2020) to minimize latency. However, it failed to enhance routing and transmission. RBM routing protocol is designed in Mukhtar et al. (2021) for enhancing network throughput with the lifetime of the network. However, routing nodes were not considered the transmission distance to save the energy further.
An efficient routing technique is designed in Aziz and Aznaoui (2020) to route data from sensor nodes to sink. However, efficient routing approaches were not considered to extend the network lifetime. FBR technique is designed in Anand et al. (2019) for minimizing computation overhead and increasing the delivery ratio. However, it failed to evaluate the performance of the FBR method with mobile node scenarios. ECRP is designed in Moussa et al. (2020) for increasing network lifetime and energy efficiency. However, the implementation of ECRP was unsuccessful in handling mobility.
An interference-aware energy-efficient routing algorithm (IA-EERA) is developed in Ding et al. (2019) to expand the network's lifetime. However, the designed algorithm failed to investigate mobility on energy-efficient routing.

Major contribution
To solve the issues reviewed from the existing literature, a novel technique called the MWIRMAB-RAR technique is introduced with a novel contribution.

Paper organization
The article is organized as follows. Section 2 describes the review of numerous energy-aware routing schemes. Section 3 presents a description of the proposed MWIRMAB-RAR technique. Section 4 presents simulation settings of proposed and existing routing schemes. Section 5 explains the performance evaluation of the MWIRMAB-RAR technique compared with the existing method. Section 6 describes the conclusion of the article.

Related works
Energy-efficient routing protocol based on reinforcement learning (ERP-RL) was introduced in Mutombo et al.
(2019) to improve the network lifetime and scalability. However, it failed to select the more optimal routing protocol to improve reliable data transmission. A novel energy-aware and reliable routing protocol was introduced in Almazaideh and Levendovszky (2020) for packet transmission. However, the designed routing technique consumes a higher degree of complexity. A multi-objective fractional particle lion algorithm is developed in Bhardwaj and Kumar (2019) to improve efficient data transmission. But, the performance of the data delivery ratio is not considered. Energy-efficient routing protocol was designed in Hadi et al. (2021) to improve the transmission framework. However, it was unsuccessful in enhancing better data transmission with reduced delay. Load adaptive beaconing scheduling (LABS) algorithm was introduced in Mohiddin and Sinha (2021) to increase the optimum network throughput. The designed algorithm failed to achieve an efficient network lifetime.
RCER protocol is developed in Haseeb et al. (2019) to improve network lifetime with lesser routing cost. However, the designed protocol failed to measure network lifetime performance. The destination oriented routing algorithm was introduced in Wang et al. (2021) with optimum distance and direction for enhancing the lifespan. But, it was unsuccessful in transmitting packets with lesser delay.
To minimize energy utilization and attain better load balancing of data transmission, energy-efficient scalable routing algorithm was designed (Elsmany et al. 2019). But, a higher delivery ratio was not achieved. Multi-objectivebased clustering and SFO-guided routing approach were introduced in Mehta and Saxena (2020) to improve energy efficiency. The designed routing algorithm failed to estimate the performance analysis of data transmission delay.
An energy-efficient region source routing protocol is introduced in Xu et al. (2019) to enhance data delivery and minimize the delay. However, the designed protocol failed to provide the theoretical analysis of network lifetime maximization.

Methodology
WSN includes several sensor devices that are developed in environmental areas for remote sensing and monitoring various conditions. WSN is referred to as a gathering of the self-organized device deployed in an unexpected way depending on ad hoc infrastructure for collection of data over network field. But resource-efficient routing in WSN is a challenging issue. Based on the motivation, a novel technique called the MWIRMAB-RAR technique is introduced.

Network model
The network model of MWIRMAB-RAR architecture was discussed. In WSN, a number of sensor nodes Si = NS1, NS2, NS3…NSn are randomly deployed in a squared grid structure 'm * m' with respect to the specific transmission range 'Rt' to collect the information from the environment. The collected information is transferred into the sink node in terms of data packets 'Pj = P1, P2,…,Pm through the energy-efficient intermediate neighboring nodes' IN1, IN2, …,INn to extend the network lifetime and select route paths for routing.
As given in the above Fig. 1, three different processes of the MWIRMAB-RAR technique are performed, such as resource-efficient node selection, route path discovery, and data transmission. A detailed description of the proposed MWIRMAB-RAR technique is provided in the forthcoming subsections.

Multivariate weighted isotonic regressive modest adaptive boosting
The proposed MWIRMAB-RAR technique starts detecting sensor nodes to improve network lifetime. The MWIR-MAB-RAR technique uses a multivariate weighted isotonic regressive modest adaptive boosting technique for identifying resource-efficient sensor nodes from the network. The proposed boosting technique is the ensemble classification technique that combines weak learners into strong learner. A weak learner is defined as a classifier that is difficult to offer true classification. A strong learner is defined as a classifier that offers well-correlated accurate classification results. Figure 2 illustrates the multivariate weighted isotonic regressive modest adaptive boosting algorithm. The ensemble boost technique considers the training samples as input, i.e., the number of sensor nodes. With training sensor nodes, an empty set of weak learners Z1, Z2, Z3,….Zk is constructed. Ensemble technique establishes various weak learners as multivariate weighted isotonic regression to detect resource-efficient sensor nodes depend on energy and bandwidth, The multivariate weighted isotonic regression is a machine learning technique for detecting relationships between outcome variables and resources. Now, the regression function determines the resources of the sensor node. In WSN, energy is a significant parameter for extending the network lifetime for each sensor node. At first, all the sensor nodes are distributed with a similar energy level. Due to the sensing and monitoring behaviors of the sensor nodes, the initial energy level is degraded. It is calculated below, In Eq. (1), encode specifies the energy of sensor nodes, p indicates power, and t stands for time. The energy of each sensor node is estimated in joule (J). In WSN, the energy level of the sensor gets corrupted through the sensing process. It is calculated by, where RENs denotes residual energy of sensor node, TNs are the total energy of sensor nodes, and CNs represent consumed energy of sensor node. The bandwidth consumption is measured as the amount of data transmitted over a particular time.
bw ¼ Amount of dp transmitted t ð3Þ where bw denotes a Bandwidth, t indicates a time, dp indicates a data packet. The bandwidth is calculated in bits per second. After measuring energy and bandwidth, the multivariate weighted isotonic regression is applied for classifying the sensor nodes. Here, the multivariate represents the two objective functions: higher residual energy and higher bandwidth. Weighted Isotonic Regression is used to estimate the variance between the objective functions and the sensor nodes.
where Y denotes weak classification results, a i denotes a weight, NS i denotes a sensor node, and f(NS) indicates an objective function of the sensor nodes. The node closer to the objective function is classified as resource-efficient sensor nodes. Otherwise, the sensor nodes are classified as normal nodes, but weak learner has some training errors. Therefore, the ensemble technique accurately classifies the sensor nodes through integrating weak learners' results.
Ensemble classification results are obtained as given below, From (5), Q symbolizes the output of the ensemble weak learner, and Y j specifies classification results of weak learners. For each weak learner's results, the weight gets initialized for identifying the strong classification results. The strong classification output of weak learners is estimated as given below, After the initialization, the generalization error of a weak learner is estimated to show how accurately an algorithm is able to provide outcome values. The error is measured for each weak learner as given below, where E denotes a generalization error, Q i denotes actual outcomes of the weak learner results, and Q a symbolizes the observed classification of weak learner results. Behind classification results, initial weight is efficient. Update weights of the classification results increase, if classified wrongly by this classifier. Otherwise, the weights of the classification results get decreased. The gradient steepest descent function finds the weak classification results with minimum generalization error.
where F denotes a gradient steepest descent function, argmin denotes an argument of the minimum function, and E (Y j ) denotes an error of weak learner results. Based on classification results, the resource-efficient sensor nodes are correctly classified.

Route path discovery
Behind finding resource-efficient sensor nodes, multiple routes are established between the source and sink nodes. Route path discovery is performed by finding the nearest neighboring node via beacon message allocation. The time of flight method is applied for measuring the distance between a sensor and an adjacent node through the beacon message distribution. At first, the source node sends a beacon message to other resource-efficient sensor nodes in the network. Then, the neighboring node sends a reply message to the source node. Figure 3 illustrates multiple route path discoveries among source and sink nodes to route data packets. At first, the source node sends a request beacon message to other sensor nodes in the network. After receiving the request beacon message, the reply is sent to the source node. Based on message arrival time, the nearest neighboring nodes are correctly identified. Therefore, the time of flight is applied as given below, where D r denotes distance, tBMA indicates the time of request beacon message arrival, and tBMT denotes the time of request beacon message transmitted. A node that sends a reply message and a lesser time is selected as the nearest neighboring node. The route path with lesser distance and minimum hop is selected for efficient data delivery. The algorithmic process of the MWIRMAB-RAR technique is described as given below.
Algorithm 1 explains the process of resource-aware routing in WSN. The proposed technique uses the ensemble classification technique for identifying the resourceefficient sensor nodes. The ensemble technique constructs the number of weak learners. For each sensor node, the energy consumption and bandwidth are measured. Then, the regression is applied for measuring the relationship between the sensor nodes and the objective function. The Multivariate weighted isotonic regressive modest adaptive boosting-based resource-aware routing in… 4943 node closer to the objective function is classified as resource-efficient sensor nodes. The weak classifications results are summed in to make a strong one. After integrating a weak learner, weight is initialized and followed by a generalization error which is expressed for every weak learner. Depending on the error value, the initial weight gets efficient. Finally, the gradient steepest function is applied for finding weak learner results with lesser error. After finding resource-efficient sensor nodes, multiple route paths among source and destination are recognized via the time of flight method. Finally, the source node performs efficient data transmission with minimum delay.

Simulation scenario
The simulation of the MWIRMAB-RAR method and existing methods, namely E3AF (El-Fouly and Ramadan 2020) and EERH (Hung et al. 2020), is developed using the NS2.34 simulator. In WSN, 500 sensor nodes are deployed in a squared area (1100 m * 1100 m). The random waypoint model is used as a mobility model. Sensor nodes are moved in network with a speed of 0 to 20 m/s. Time is 300 s. In WSN, DSR protocol is used for resource-efficient routing. Table 1 shows simulation parameters are listed.

Performance analysis
The proposed MWIRMAB-RAR technique and existing methods E3AF (El-Fouly and Ramadan 2020) and EERH (Hung et al. 2020) were discussed with different performance metrics. The performance of different metrics is evaluated with table and graphical representation to determine the network scalability through varying sensor nodes.
• Energy consumption is a significant metric to extend network lifetime. It is calculated as the amount of energy consumed by sensor nodes to deliver data. It is formulated as follows, where Comp ER represents energy consumption, NS i indicates a number of sensor nodes, and SN' denotes single sensor nodes. It is measured in joule (J).
• Packet delivery ratio: It is a major routing metric used to demonstrate the proposed MWIRMAB-RAR method. It is measured as the proportion of packets that are effectively received at the sink node to the entire amount of packets sent. Therefore, the overall delivery ratio is measured by, where DP r denotes packet delivery ratio, NPD indicates the number of packets delivered, and NPS denotes the amount of packet sent. It is calculated in percentage (%) • Packet loss rate: Another routing metric is measured as the proportion of packets lost at the sink node to the entire number of packets sent. It is estimated as given below, where DP L denotes a packet loss rate, NPL denotes the number of packets lost, and NPS denotes the number of packet sent. It is measured in percentage (%).
• End-to-end delay: It is calculated as the difference between the expected arrival time of data and the actual arrival time of the packet at the sink node. Delay of data delivery from source to sink is calculated as given below, where 'ED' indicates an end-to-end delay, tact designates an actual packet arrival time, and 't Ex ' indicates expected arrival time. It is measured in milliseconds (ms). Table 2 indicates the simulation analysis of energy consumption with the amount of mobile nodes. The amount of mobile nodes is taken in the ranges from 50, 100, 150, Source Sink Request beacon message Reply beacon message Fig. 3 Route path discovery 200…500 for ten iterations with three different routing methods MWIRMAB-RAR technique and existing methods, E3AF (El-Fouly and Ramadan 2020) and EERH (Hung et al. 2020). For each routing technique, ten different results are observed. The obtained results indicate that the MWIRMAB-RAR technique provides superior performance than the existing methods. Let us consider 50 sensor nodes for performing the simulation. The energy consumption of the MWIRMAB-RAR technique was observed 17 J, and the performances of energy consumption observed using E3AF (El-Fouly and Ramadan 2020) and EERH (Hung et al. 2020) are 20 J, 22 J, respectively. For each routing technique, ten different results are observed. The performance outcomes of energy consumption of the MWIRMAB-RAR technique are compared to existing methods. The compassion of ten outcomes indicates that the energy consumption of the proposed MWIRMAB-RAR method is significantly reduced by 12% compared with E3AF (El-Fouly and Ramadan 2020) and 21% compared with EERH (Hung et al. 2020). Figure 4 illustrates energy consumption with the amount of sensor nods. From Fig. 4, the amount of sensor nodes (along y-axis) and energy consumption in terms of joule (along with the y-axis) are related to the extent of the network of the WSN. The observed results indicate that the energy consumption for all the methods gets increased. But comparatively, the MWIRMAB-RAR technique outperforms well than other routing schemes. This improvement is archived through the application of finding the resourceefficient sensor nodes for data delivery. Resource-efficient sensor nodes are identified through the multivariate weighted isotonic regressive modest adaptive boosting. The ensemble method is employed for finding better energy-efficient nodes for transmitting data. As a result, the overall energy consumption of the MWIRMAB-RAR technique is minimized than the existing methods. Table 3 reports the performance results of packet delivery ratio versus amount of packets being sent. The number of data considered for simulation is varied from 50, 100 …500. The table value indicates that the performance of packet delivery ratio of three routing techniques, MWIRMAB-RAR technique, E3AF (El-Fouly and Ramadan 2020) and EERH (Hung et al. 2020) is obtained. Among three methods, the MWIRMAB-RAR technique outperforms well in achieving high data delivery. With consideration of 50 data packets being sent, the amount of successful packets sent at the destination is 45 and packet delivery ratio of the MWIRMAB-RAR technique is 90%. Similarly, the packet delivery ratio of E3AF (El-Fouly and Ramadan 2020) and EERH (Hung et al. 2020) is 88%, 86%, respectively. Followed by nine results are obtained for each routing technique. The observed performance results of the MWIRMAB-RAR technique are compared to existing methods. The average of ten comparison results is taken into consideration of final results. The comparison results provide the overall packet delivery ratio of the MWIRMAB-RAR technique that is considerably increased by 4% and 7% than the existing (El-Fouly and Ramadan 2020) (Hung et al. 2020).  50,100,150,200,250,300,350,400,450,500 Number of data packets 50,100,150,200,250,300,350,400,450,500 Mobility Multivariate weighted isotonic regressive modest adaptive boosting-based resource-aware routing in… 4945 Figure 5 explains the packet delivery ratio of three routing techniques. The packet delivery ratio of y-axis is represented with amount of data packets on x-axis. As shown in Fig. 5, the delivery ratio of three different methods, MWIRMAB-RAR technique, E3AF (El-Fouly and Ramadan 2020), and EERH (Hung et al. 2020), is represented by three different colors, namely blue, red, and green. The graphical plot indicates that the MWIRMAB-RAR technique provides superior performance than the other two methods. This is because the MWIRMAB-RAR technique finds resource-efficient nodes for the routing process. The higher energy-efficient nodes and higher bandwidth increase the data transmission. As a result, a successful data delivery ratio is obtained at the sink node.
The comparative result analysis of the packet loss rate is depicted in Table 4. The MWIRMAB-RAR technique is implemented in the NS2 simulator with a varied number of data packets in the range of 50-500 for calculating the packet loss rate. When conducting the simulation with 50 data packets in the first iteration, the proposed MWIRMAB-RAR technique lost 5 data packets, whereas E3AF (El-Fouly and Ramadan 2020) and EERH (Hung et al. 2020) lost 6 and 7 data packets, respectively. As a result, the packet loss rate of the MWIRMAB-RAR technique is 10%, and the loss rate of existing (El-Fouly and Ramadan 2020; Hung et al. 2020) is 12% and 14%, respectively. The proposed technique is compared with existing methods. The overall comparison results indicate that the packet loss rate of the MWIRMAB-RAR method decreased as 31% compared with E3AF (El-Fouly and Ramadan 2020) and 42% compared with EERH (Hung et al. 2020). Figure 6 presents the impact of packet loss rate with a diverse amount of data with proposed and two existing methods. The graphical diagram shows that MWIRMAB-RAR method decreases the packet loss rate than the other two conventional routing schemes. This is because the application of the node which has better bandwidth and higher energy efficiency is selected to route data. This minimizes data loss and increases data delivery.    Table 5 and Fig. 7 present delay along with a diverse number of packets using three methods. The graphical result of the proposed MWIRMAB-RAR method offers a lesser amount of delay in order to accurately perform the data transmission when compared to conventional E3AF (El-Fouly and Ramadan 2020) and EERH (Hung et al. 2020). Let us consider '50' data sent, delay of MWIR-MAB-RAR technique was found to be '12 ms,' and the delay of existing E3AF (El-Fouly and Ramadan 2020) and EERH (Hung et al. 2020) is '14 ms' and '16 ms,' respectively. The above estimated mathematical result confirms that the MWIRMAB-RAR technique outperforms well in minimizing delay. The overall assessment of ten outcomes confirms delay for delivering data is decreased by 11% compared with (El-Fouly and Ramadan 2020) and 19% compared with (Hung et al. 2020), respectively. The main motivation of enhancement is used to find energy and bandwidth. In addition, the MWIRMAB-RAR technique finds the nearest neighboring node via a time of flight method. Then, the MWIRMAB-RAR technique discovers the route path with minimum distance for enhancing data delivery and reducing delay.

Conclusion
This paper innovatively applies the new technique called MWIRMAB-RAR which is introduced for enhancing the energy and reliability of WSN. The MWIRMAB-RAR technique finds the resource-efficient sensor nodes by applying the modest adaptive boosting technique. The ensemble technique uses the weak classifier for finding the efficient sensor nodes based on energy and bandwidth. After selecting the resource-efficient sensor node, the route paths are constructed based on the time of flight method for identifying the neighboring sensor node. Finally, the data transmission is performed through the resource-efficient nodes. This increases data delivery and decreases packet loss and delay. The simulation investigation is carried out on certain performance factors. The proposed MWIR-MAB-RAR method offers a high delivery rate, decreases loss rate, delay, and decreases energy consumption than conventional methods.