1
An Incremental Learning Classification Algorithm
based on Forgetting Factor for eHealth Networks
Li Yang
1
, Kun Wang
1
, Chenhan Xu
1
, Chunsheng Zhu
2
, Yanfei Sun
1
1
Key Lab of Broadband Wireless Communication and Sensor Network Technology, Ministry of Education,
Nanjing University of Posts and Telecommunications, China.
Emails: islyang@foxmail.com, kwang@njupt.edu.cn, xchank@outlook.com, njsyf@vip.163.com
2
Department of Electrical and Computer Engineering, The University of British Columbia, Canada.
Email: cszhu@ece.ubc.ca
Abstract—The advances of network technology and mobile
communication technology are making eHealth possible. In
eHealth systems, physiological data and relevant context-aware
data are acquired continuously and in real time. At the same time,
such large-scale data results in huge challenges in the aspect of
real-time big data processing since eHealth data appears in the
form of data stream. Therefore, we propose a novel incremental
learning algorithm, namely α-SVMSGD, which improves the
SVMSGD (Support Vector Machine-Stochastic Gradient Descent)
algorithm by updating the training data with the continuous data
stream. Besides, this α-SVMSGD may handle the problem that
original SVMSGD cannot further mine the useful information
in unclassified data. In α-SVMSGD, the process of training data
updating is completed by introducing the concept of forgetting
mechanism, in which the forgetting factor α is introduced to weed
out useless training data. α-SVMSGD is applied into ambient
assisted living communications, and further incorporated into
the data filtering layer of a local data processing architecture
(LDPA) to reduce data redundancy. Simulation results confirm
that the proposed algorithm is a promising data redundancy
solution for classification without loss of accuracy in the case of
real-time data stream.
Keywords—eHealth, Incremental learning, Support vector ma-
chine, Stochastic gradient descent, Forgetting factor
I. INTRODUCTION
Due to the hospital capacity and medical staff are limited
concerning the increasing treatment requests, traditional health
care services can hardly satisfy growing population’s needs.
Under this background, for the benefits of big data technique,
a new kind of eHealth service monitoring people’s lives with
intelligent device is developing rapidly. In this current era
of big data [1], the Internet transmits a great deal of data,
followed with the data storage and data processing by servers
or clouds. Besides, mobile networks collect lots of data all
around people’s lives. Due to the limitations of traditional
data processing methods, they are often used to describe those
complex or large data sets in the network.
Specifically, the growing amount of data collected by mobile
eHealth networks is more and more pervasive with the devel-
opment of hardware. Meanwhile, collecting nodes are required
to get more data [2] and the number of them tends to increase
in networks. All the above factors increase the scale of network
and the volume of data transferred in the e-Health network.
Since the energy and functionality of mobile nodes are
limited, data has to be aggregated and processed in a central
sever. However, with the increment of network size which
leads to the emergence of more powerful functions, it is likely
to cause that the central server is not capable of analyzing all
the data due to various factors (e.g., routing blocking resulting
from malicious nodes or network congestions). For this reason,
how to efficiently process these data is a very important
problem. In our previous work, we proposed the framework of
a local data processing architecture (LDPA) to provide the idea
of quantifying the result of data analysis in ambient assisted
living communications (AAL) [3]. However, data redundancy
issue still exists in data filtering layer (DFL) of LDPA. To deal
with this problem, this paper utilizes an improved SVMSGD
(Support Vector Machine-Stochastic Gradient Descent) [4]
algorithm by introducing the concept of forgetting factor α
to update training data. Besides, in original SVMSGD, some
past information in classifier may be new classifications, and
SVMSGD cannot further process this kind of data. We then
propose an incremental learning algorithm α-SVMSGD and
further incorporate it into data filtering layer (DFL) of LDPA
for mining useful information from unclassified data.
To this end, we mainly focus on efficient data processing
and mining useful data classification from the original data and
increasing data to save storage space and reduce the probability
of data loss due to network congestion. The contributions of
our work are summarized as follows:
• A forgetting factor method is proposed to mine useful
information from the unclassified data. It can mine new
useful information from these useless or incremental
data, and reduce history sample storage and the scale of
training sample set.
• In α-SVMSGD, we adopt adaptive method to adjust the
value of α. We set a threshold for α firstly, and then
calculate the error between initial threshold and α value
of sample which has been training for several times.
Finally, we choose α with maximum error weights as
a new threshold and begin the next round of the training
of new data sample.
The rest of paper is organized as follows. Section II reviews
related work. Section III presents the overview of LDPA.