Privacy-Preserving Electrocardiogram Monitoring for Intelligent Arrhythmia Detection

View full-text article in PMC

. 2017 Jun 12;17(6):1360. doi: 10.3390/s17061360

Algorithm 1 C4.5 algorithm for arrhythmia learning.

1:
function DecisionTree(D, A, T) ▷ D: training dataset, A: feature set, T: decision tree
2:
if D contains only training examples of the same heartbeat type $c_{j} \in C$ then
3:
Make T a leaf node labeled with heartbeat type $c_{j}$ ; ▷ C: heartbeat types
4:
else if $A = Ø$ then
5:
Make T a leaf node labeled with heartbeat type $c_{j}$ , which is the most frequent heartbeat type in D;
6:
else ▷ D contains examples belonging to a mixture of heartbeat types
7:
$p_{i}$ =impurityEval- $1 (D)$ ; ▷ We select a single feature to partition D into subsets so that each subset purer
8:
for Each feature $A_{i} \in A (= {A_{1}, A_{2}, \dots, A_{k}})$ do
9:
$p_{i}$ =impurityEval- $2 (A_{i}, D)$ ;
10:
end for
11:
Select $A_{g} \in {A_{1}, A_{2}, \dots, A_{k}}$ that provides the biggest impurity reduction, computed using $p_{0} - p_{i}$ ;
12:
if $(p_{0} - p_{g}) < threshold$ then ▷ $A_{g}$ does not significantly reduce impurity $p_{0}$
13:
Make T a leaf node labeled with $c_{j}$ , the most frequent heartbeat type in D;
14:
else ▷ $A_{g}$ is able to reduce impurity $p_{0}$
15:
Make T a decision node on $A_{g}$ ;
16:
Let the possible values of $A_{g}$ be $v_{1}, v_{2}, \dots, v_{m}$ . Partition D into m disjoint subsets $D_{1}, D_{2}, \dots, D_{m}$ based on the m values of $A_{g}$ ;
17:
for Each $D_{j} \in D (= {D_{1}, D_{2}, \dots, D_{m}})$ do
18:
if $D_{j} \neq Ø$ then
19:
Create a branch (edge) node $T_{j}$ for $v_{j}$ as a child node of T;
20:
DecisionTree $(D_{j}, A - {A_{g}}, T_{j})$ ;
21:
end if
22:
end for
23:
end if
24:
end if
25:
end function