Page 253 - ITU Kaleidoscope 2016

ICTs for a Sustainable World

expert system knowledge based algorithm which learns from both individuals with 1) normal vital signs indication accord-
the data, calculates the weights for each variable or generates ing to the WHO norms during the four days and 2) very little
a linear hypothesis which it uses to score the vital parameters. daily variations since the users did not fall sick during the
experimentation. These values were calibrated against those
obtained from medical equipment used by nurses in hospi-
3.2. K-means Clustering Algorithm
tals and bench-marked against the WHO values[1]. They re-
The K-means clustering (KMC) algorithm considered in this vealed similar values and performance patterns for non-sick
paper partitions n observations into k sets (k < n) S = individuals.
{S 1 , S 2 , . . . , S k } so as to minimize the within-cluster sum
of errors squares (WCSS), which is expressed by:
4.2. Information Dissemination
k
X X 2
arg s min || x j − µ i || We conducted a set of experiments using a worst case de-
ployment scenario for a rural hospital represented by an over-
i=1 x j ∈S i
crowed building complex in Cape Town with many tenants
where µ i is the mean of points in S i . using WiFi devices (laptops, tablets and phones) to access
the Internet and communicate through social media. The ex-
perimental results are depicted by Figures 6 and 7 for the
RSSI and throughput respectively. The signal strength at the
receiver’s side for the IEEE 802.11 communication is con-
stant over different distances while the signal strength in the
802.15.4 links decreases with distance. This is in line with
the fact that the 802.11 protocol has been designed for longer
communication ranges than the 802.15.4 protocol. Further-
more, the IEEE 802.11, basis of WiFi, reveals similar re-
sults for both indoor and outdoor communication. In con-
Figure 5: K-means clustering
trast, the 802.15.4 shows a difference between indoor and
outdoor scenarios where outdoor links reach longer ranges
Algorithm description. As described by Figure 5, the KMC than indoor links and the indoor RSSI strength reduces with
algorithm considered in this paper partition the data into clus- the number of walls. To measure the throughput achieved
ters and uses the Gaussian estimator (Parzen window estima- over indoor and outdoor communication links, we transmit-
tor) algorithm to estimate a probability density function p(x) ted a number of packets configured to contain the bio-sensor
correspondent to a particular status which is then used to cal- readings as payload and measured the ratio of the number of
culate a patient status index (PSI) by means of the equation packet successfully received and acknowledged to the num-
PSI = log e [1/p(x)]. ber of packets sent. Figure 7 reveals a performance pat-
tern similar to the received signal strength indicator (RSSI)
where the throughput achieved indoor and outdoor are the
4. PERFORMANCE EVALUATION
same for the 802.11 protocol. The 802.15.4 reveals differ-
ent performance patterns between indoor and outdoor com-
The Cyber-healthcare system used in our work leverages the
munication and for the indoor communication with differ-
off-the-shelf e-Health kit from Libelium as a low cost de-
ent number of walls. Thanks to the high RSSI, a constant
vice that can be easily and quickly deployed in a rural en-
percentage of packets was received for the 802.11 protocol
vironment. The development environment used in our ex-
while the indoor 802.15.4 links achieved a higher through-
periments included: Ubuntu 13.10, Android SDK, Apache2
put for one wall compared to two walls. The 802.11 pro-
server (on Ubuntu), MySQL server, MySQL -Java-Bridge
tocol achieved higher throughput compared to the 802.15.4
(MySQL-J-Connector). Different languages were used dur-
in both indoor and outdoor communications. Note that al-
ing our development, including Java, PHP, SQL, JSON, and
though the 802.11 protocol outperformed the the 802.15.4
XML.
on both performance parameters, the 802.15.4 deployment is
a cheaper option and more frugal option in terms of energy
4.1. Sensor Field Readiness consumption even when using the lightweight version of the
IEEE802.11 protocol often referred to as WiFi-lite.
We conducted a first set of experiments to evaluate the field
readiness of the off-the-shelf e-health sensor technology with
the objective of making sure that the sensor readings fall in 4.3. Health Kiosks Mesh Network Engineering
acceptable ranges. To overcome the lengthy ethical clear-
ance procedures aimed at protecting patient privacy through The information dissemination experiments presented above
confidentiality, we used in this experiment two healthy users were complemented by a network engineering study of the
whose vital signs were monitored for four days. The results feasibility of a 45 nodes community health care network in
presented in tables 1 confirmed a normal healthy state for Lubumbashi as described earlier. The results produced using

– 235 –