Volume 7, Issue 3, March – 2022 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Handoff using Machine Learning Techniques

P. Sai Krishna, K. Sai Harsha Vardhan Reddy, K. Srikar DR. Rama Swamy, Dr. Gobinda Prasad Acharya
Student, Department of ECE Associate Professor, Department of ECE,
Sreenidhi Institute of Science and Technology Sreenidhi Institute of Science and Technology
Hyderabad, India. Hyderabad, India.

Abstract:- This paper demonstrates about

Implementation of Handoff Techniques through Machine
Learning Algorithms in Tele communications. A handoff
is the process of transferring an active call or data session
from one cell in a cellular network to another, or from
one channel within a cell to another. Cellular networks
are made up of cells, each of which can provide
telecommunications services to customers roaming
through the network. Each cell has a limited region and
number of subscribers it can serve. A handoff occurs
when any of these two thresholds is reached. When a
certain mobile tower's capacity is exceeded, an existing or
new call from a phone must be transferred to another cell
tower that covers the same geographical area as the
existing cell tower. A well-executed handoff is essential
for providing continuous service to a caller or data
session user. Using Machine learning algorithms. The Fig. 1: Describes the NM and the switch networks
present methodology examines the accuracies generated
by three popular decision-making algorithms namely II. RELATED WORK
Logistic Regression, Decision trees, Random Forests.
Handoff has been an important research problem because
Keywords:- Handoff, roaming, cellular networks, Logistic of its core problem of decreasing delay and associated
Regression, Decision trees, Random Forests. transitions. Handoff, the overall procedure of the transition
process, has been divided into three steps. 1) Measurement and
I. INTRODUCTION preparation for the handover, 2) decision for the handover, and
3) execution of the handover. In the first stage, metrics from
The design of an efficient Handoff Management is one the next networks are measured; in the second stage,
of the most difficult challenges for improving quality of algorithms determine when to switch networks; and in the
service (QoS) in wireless networks. It is necessary to assist a
third stage, the necessary procedures to connect to the new
mobile terminal's decision to switch between different types
network and reestablish services are carried out. In the
of networks. In this paper we present a machine learning
presence of barriers, a machine learning-based handover
approach to improve the quality of service(QOS) by
management system for LTE has been employed to improve
considering various contributing factors. A Neural Network- the user's Quality of Experience. The typical Wireless Local
based model assists the Handoff device in making the most Area Network (WLAN) handoff management system causes
efficient Handoff decision. The simulation findings suggest visible delays during the handoff procedure, resulting in
that using a neural network-based machine learning method service interruption, which is especially noticeable in
to carry out the Handover process can improve the QoS congested WLANs. The RNN-HM system is a revolutionary
experienced by both voice and data services while also handoff management technique based on deep learning,
meeting the user's preferences to a large extent. In IP specifically the recurrent neural network (RNN). RNN-HM
networks, the handover or handoff is the physical transfer may effectively improve the data rate during the handoff
from one network to another. There are two sorts of process, exceeding the standard scheme, according to
transitions in the handover. 1) horizontal handover and 2)
numerical results obtained through simulation [1].
vertical handover A horizontal handoff occurs when a Node
Mobile (NM) switches networks in the same technology. As Handover methods are complicated by a number of
seen in the dotted box of Figure 1, if the network switches to aspects, some of which can make quantification difficult.
a new technology, a vertical handoff is required. As depicted Fuzzy logic is being used to handle change decision problems
in Figure that include fuzzy parameters [2]. The fuzzy logic-based
approach allows for systematic tweaking of the handoff
settings, resulting in a fair tradeoff between various system
features [3]. The information on the parameters linked to
handover decisions in wireless communication networks, as
well as their operating range, was collected in fuzzy logic for
creating handover decision support systems. The table below
shows the parameters along with their working ranges [4].

IJISRT22MAR537 www.ijisrt.com 613

Volume 7, Issue 3, March – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
III. MODEL AND TOOLS USED C. Random Forest:
Random Forest is a well-known machine learning
A. Logistic regression: algorithm that uses the supervised learning method. In
Logistic regression(Multivariate), has been employed for machine learning, it can be utilized for both classification and
making handoff decisions based on the simulation data regression issues. It is based on ensemble learning, which is a
generated. Fig2 depicts the model which is used for model method of integrating several classifiers to solve a complex
training, the machine learning model has parameters SNR, problem and increase the model's performance.fig 4 depicts
RSS, NwCOv, PowCon, SNR which are used for training the the random forest analogy.
model for over 800 samples, the model has a sigmoid
function which takes care of the decision handling process.
And fig 3 shows how the model is analogous to the present
problem of handoff.

Fig. 4: Random forest analogy

D. Handoff Decision
The decision-making algorithm will have a direct impact
on the performance of the handoff. The algorithm proposed in
the project employs schemes based on Received Signal
Strength (RSS), Quality of Service (QoS), multi criteria
Fig. 2: Shows the model parameters and sigmoid function decision functions, and algorithms based on artificial
for decision making. intelligence techniques. In general, the decision-making
algorithm is fed data from the network and, after processing
the data, decides which network to change. Received Signal
Strength Indicator (RSSI), power level of signals received in
wireless networks and Network Load are some of the most
common criteria used to make the decision to change.

IV. PARAMETERS CONSIDERED AND THEIR

WORKING RANGE

Parameter Parameter Functions Favourable

Considered Ranges conditions
for handoff

Received signal -100 to -20 Weak, medium, weak

Fig. 3: depicts the analogy of how model is employed
strength(RSS) dB strong
B. Decision Tree:
A decision tree is a classification method that splits the Band-width (Bw) 1kbps to low, average, low
instance space in a cyclical fashion. The nodes inside the 20MHZ high
decision tree form a rooted tree, which is a directed tree with
no incoming edges and a "root" node. There is one incoming Network 0-72Km Near, medium, far
edge for each of the other nodes. A node having outgoing coverage far
edges is known as an internal or test node. The term "leaf" (NwCov)
refers to any additional nodes (also known as terminal or
decision nodes). Based on a discrete function of the input Power 0-2 Watts Less,medium,hig high
attribute values, each internal node in a decision tree divides consumption h
the instance space into two or more sub-spaces. In the (PowCon)
simplest and most frequent scenario, each test evaluates a
single attribute, with the instance space partitioned according Signal to Noise 20-50 db Weak, medium, high
to that same value of the attribute In the case of numeric Ratio(Snr) High
properties, the condition refers to a range. A class is assigned
to each leaf based on the best target value. Table 1: Approximated data ranges for parameters

IJISRT22MAR537 www.ijisrt.com 614

Volume 7, Issue 3, March – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
V. RESULTS

The training is done for 600 samples and the confusion

matrix is plotted over testing data. Fig below shows TPs (True
positives), FPs (False positives), TNs (True negatives), FNs
(False negatives), and corresponding accuracy scores,
precision scores are depicted. The confusion matrix is plotted
for 400 data points out of which 174 are true negatives and
170 are true positives. 25 are false positives and 31 are false
positives in case of logistic regression. The model has done a
decent job in predicting if a handoff has to be made to a
particular station. The model had to be tested and tuned
Fig. 5: depicts a sample of data generated and associated according to the real-world case scenarios for best fit. Fig 8
label of handoff decision. shows the same

The decision tree model has also been implemented

working of which is explained in the above diagrams, the
decision tree gave out an accuracy of 83%, which accounts to
single decision tree usage which shall affect the model. In total
the true positives generated are 181 and true negatives
generated are 151, and false positives and false negatives
account for 38 and 30 respectively. fig 9 depicts the confusion
matrix for the same.

The Random Forest algorithm trained with the same data

gave an accuracy of 87.5 which is considered as one of the
best algorithms in the case of handoff decision out of selected
Fig. 6: statistical analysis of the data generated algorithms of interest. Fig10 depicts the confusion matix for
the same.

Fig. 7: correlation between parameters and handoff decision

Experiments have been carried out with parameters

such as received signal strength, bandwidth, power
consumption, and network coverage, Signal to Noise Ratio
and corresponding ranges have been concluded as in table 1
The data is collected according to the approximations from
the table and from the local service provider. Among all the
parameters considered, Network Coverage is given utmost
priority, while Power consumption is given the lowest.fig 5 Fig. 8: confusion matrix for the predicted label using Logistic
shows the sample data generated with simulation regression
environments. Fig6 shows the statistical analysis of the data
that is generated.Fig7 shows the correlation between data that
is helpful in analysing parameters and concluding what all
the parameters are significant in making a decision of
handoff. The entire data has been imported into the python
environment for comprehensive evaluation with models
undertaken.

IJISRT22MAR537 www.ijisrt.com 615

Volume 7, Issue 3, March – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165

Algorithm Used Accuracy achieved

Logistic regression 86%

Decision tree 83%

Random forest 87%

REFERENCES

[1.]Han, Zijun, et al. "Artificial intelligence based handoff

management for dense WLANs: A deep learning
approach." 2018 IEEE Globecom Workshops (GC
Wkshps). IEEE, 2018.
[2.]Bazán, Josué Vicente Cervantes, et al. "Artificial
intelligence techniques in handover decision: A brief
review." Revista Ingeniantes 6.1 (2019): 1.
[3.]Patil, Chandrashekhar G., and Mahesh T. Kolte. "An
Fig. 9: predicted labels using Decision trees approach for optimization of the handoff algorithm using
fuzzy logic system." International Journal of Computer
Science and Communication 2.1 (2011): 113-118.
[4.]Payal Mahajan, Zaheeruddin,Handoff Decisions in Hetnets
using Fuzzy Logic and Machine Learning
Techniques,International Journal of Engineering and
Advanced Technology (IJEAT),ISSN: 2249 – 8958,
Volume-9 Issue-1, October 2019

Fig. 10: predicted labels using Random forest

VI. CONCLUSIONS AND FUTURE WORK

In this comprehensive analysis of machine learning

techniques for handover decisions, the logistic regression,
Decision trees, and Random Forests technique have used
1000 samples of the dataset which is created through
simulation to make handover decisions. The training data was
divided into three categories: training (600 samples), test
(200 samples), and validation (200 samples) (200 samples).
This approach yields an accuracy of 86% of accuracy for
logistic regression, 83% of accuracy for Decision trees and
87% for Random forests. It can be concluded from the
simulations data that the Random Forest algorithm has given
the most accuracy for handoff decisions, The present solution
aims to consider various devices doing the handover process
at the same time in future development.

IJISRT22MAR537 www.ijisrt.com 616