Decision Tree Classifiers to determine the patient’s Post-operative Recovery Decision

D.Shanthi, Dr.G.Sahoo & Dr.N.Saravanan
International Journal of Artificial Intelligence and Expert Systems (IJAE), Volume (1): Issue (4) 75
Decision Tree Classifiers to Determine the Patient’s Post-
Operative Recovery Decision
D.Shanthi dshan71@gmail.com
Department of Computer Science and IT
Mazoon University College
Seeb, P.O.Box 101, Muscat, Sultanate of Oman
Dr.G.Sahoo gsahoo@bitmesra.ac.in
Department of Information Technology
Birla Institute of Technology
Mesra, Ranchi, India
Dr.N.Saravanan saranshanu@gmail.com
Department of Computer Science and IT
Mazoon University College
Seeb, P.O.Box 101, Muscat, Sultanate of Oman
Abstract
Machine Learning aims to generate classifying expressions simple enough to be
understood easily by the human. There are many machine learning approaches
available for classification. Among which decision tree learning is one of the most
popular classification algorithms. In this paper we propose a systematic approach
based on decision tree which is used to automatically determine the patient’s
post–operative recovery status. Decision Tree structures are constructed, using
data mining methods and then are used to classify discharge decisions.
Keywords: Data Mining, Decision Tree, Machine Learning, Post-operative Recovery.
1. INTRODUCTION
Decision support systems help physicians and also play an important role in medical decision
making. They are based on different models and the best of them are providing an explanation
together with an accurate, reliable and quick response. Clinical decision-making is a unique
process that involves the interplay between knowledge of pre-existing pathological conditions,
explicit patient information, nursing care and experiential learning [1]. One of the most popular
among machine learning approaches is decision trees. A data object, referred to as an example,
is described by a set of attributes or variables and one of the attributes describes the class that
an example belongs to and is thus called the class attribute or class variable. Other attributes are
often called independent or predictor attributes (or variables). The set of examples used to learn
the classification model is called the training data set. Tasks related to classification include
regression, which builds a model from training data to predict numerical values, and clustering,
which groups examples to form categories. Classification belongs to the category of supervised
learning, distinguished from unsupervised learning. In supervised learning, the training data
consists of pairs of input data (typically vectors), and desired outputs, while in unsupervised
learning there is no a priori output. For years they have been successfully used in many medical
decision making applications. Transparent representation of acquired knowledge and fast
algorithms made decision trees what they are today: one of the most often used symbolic
machine learning approaches [2]. Decision trees have been already successfully used in

medicine, but as in traditional statistics, some hard real world problems cannot be solved
successfully using the traditional way of induction [3]. A hybrid neuro-genetic approach has been
used for the selection of input features for the neural network and the experimental results proved
that the performance of ANN can be improved by selecting good combination of input variables
and this hybrid approach gives better average prediction accuracy than the traditional ANN [4].
Classification has various applications, such as learning from a patient database to diagnose a
disease based on the symptoms of a patient, analyzing credit card transactions to identify
fraudulent transactions, automatic recognition of letters or digits based on handwriting samples,
and distinguishing highly active compounds from inactive ones based on the structures of
compounds for drug discovery [5]. Seven algorithms are compared in the training of the multi-
layered Neural Network Architecture for the prediction of patient’s post-operative recovery area
and the best classification rates are compared [6].
In this study, the Gini Splitting algorithm is used with promising results in a crucial way and at the
same time complicated classification problem concerning the prediction of post-operative
recovery area.
2. MOTIVATIONS AND RELATED WORK
In data mining, there are three primary components: model representation, model evaluation and
search. The two basic types of search methods used in data mining consist of two components:
Parameter Search and Model Search [15]. In parameter search, the algorithm searches for the
set of parameters for a fixed model representation, which optimizes the model evaluation criteria
given the observed data. For relatively simple problems, the search is simple and the optimal
parameter estimates can be obtained in a closed form. Typically, for more general models, a
closed form solution is not available. In such cases, iterative methods, such as the gradient
descent method of back-propagation for neural networks, are commonly used. The gradient
descent method is one of the popular search techniques in conventional optimization [16].
Decision trees are considered to be one of the most popular approaches for representing
classifiers. It induces a decision tree from data. A decision tree is a tree structured prediction
model where each internal node denotes a test on an attribute, each outgoing branch represents
an outcome of the test, and each leaf node is labeled with a class or class distribution. Decision
trees are often used in classification and prediction. It is simple yet a powerful way of knowledge
representation. The models produced by decision trees are represented in the form of tree
structure. Learning a decision tree involves deciding which split to make at each node, and how
deep the tree should be. A leaf node indicates the class of the examples. The instances are
classified by sorting them down the tree from the root node to some leaf node [2].
FIGURE 1: Decision Tree Structure

Decision trees (DTs) are either univariate or multivariate [17]. Univariate decision trees (UDTs)
approximate the underlying distribution by partitioning the feature space recursively with axis-
parallel hyperplanes. The underlying function, or relationship between inputs and outputs, is
approximated by a synthesis of the hyper-rectangles generated from the partitions. Multivariate
decision trees (MDTs) have more complicated partitioning methodologies and are
computationally more expensive than UDTs. A typical decision tree learning algorithm adopts a
top-down recursive divide-and-conquer strategy to construct a decision tree. Starting from a root
node representing the whole training data, the data is split into two or more subsets based on the
values of an attribute chosen according to a splitting criterion. For each subset a child node is
created and the subset is associated with the child. The process is then separately repeated on
the data in each of the child nodes, and so on, until a termination criterion is satisfied. Many
decision tree learning algorithms exist. They differ mainly in attribute-selection criteria, such as
information gain, gain ratio [7], gini index [8], termination criteria and post-pruning strategies.
Post-pruning is a technique that removes some branches of the tree after the tree is constructed
to prevent the tree from over-fitting the training data. Representative decision tree algorithms
include CART [8] and C4.5 [7]. There are also studies on fast and scalable construction of
decision trees. Representative algorithms of such kind include RainForest [9] and SPRINT [10].
3. METHODS AND METHODOLOGY
The tool used in this study is known as DTREG [14]. It basically follows the same principles as
many other decision tree building tools, but it also implements different extensions. One of those
extensions is called dynamic discretization of continuous attributes, which was used in our
experiments with success. For this study, we have taken classification problem. The problem
consists of determining the class (General Hospital, Go-home, Intensive care) for a certain input
vector. For this study, the data was taken from UCI Machine Learning Repository.
(http://www.ics.uci.edu/~mlearn/ MLSummary.html) [11]. The data were originally created by
Sharon Summers (School of Nursing, University of Kansas) and Linda Woolery (School of
Nursing, University of Missouri) and donated by Jerzy W. Grzymala-Busse. The goal is to
determine where patients in a postoperative recovery area should be sent to next. Because
hypothermia is a significant concern after surgery. The attributes correspond roughly to body
temperature measurements. The dataset contains 90 records with 8 characteristics of a patient's
state in a postoperative period.
1) InternalTemp - patient's internal temperature in C: high (> 37), mid (>= 36 and <= 37),
low (< 36)
2) SurfaceTemp - patient's surface temperature in C: high (> 36.5), mid (>= 36.5 an d <= 35),
low (< 35)
3) OxygenSat - oxygen saturation in %: excellent (>= 98), good (>= 90 and < 98),
fair (>= 80 and < 90), poor (< 80)
4) BloodPress - last measurement of blood pressure: high (> 130/90), mid (<= 130/90
and >= 90/70), low (< 90/70)
5) SurfTempStab - stability of patient's surface temperature: stable, mod-stable, unstable
6) IntTempStab - stability of patient's internal temperature: stable, mod-stable, unstable
7) BloodPressStab - stability of patient's blood pressure: stable, mod-stable, unstable
8) Comfort - patient's perceived comfort at discharge, measured as an integer between 0 and 20
A. Target Column
9) Discharge Decision - discharge decision (Intensive-Care, Go-Home, General-Hospital):
Intensive-Care (patient sent to Intensive Care Unit),
Go-Home (patient prepared to go home),
General-Hospital (patient sent to general hospital floor).
4. EXPERIMENTAL RESULTS AND EVALUATION
The input to a classification problem is a dataset of training records and each record has several
attributes. Attributes whose domains are numerical are called numerical attributes (continuous
attributes), whereas attributes whose domains are not are called categorical attributes (discrete

attributes). There is one distinguished attribute, called class label, which is a categorical attribute
with a very small domain. The remaining attributes are called predictor attributes; they are either
continuous or discrete in nature. The goal of classification is to build a concise model of the
distribution of the class label in terms of the predictor attributes [12].
Sl.No Attributes
1. InternalTemp
2. SurfaceTemp
3. OxygenSat
4. BloodPress
5. SurfTempStab
6. IntTempStab
7. BloodPressStab
8. Comfort
TABLE 1: Input Attributes
In this data analysis the last column will be considered as the target one and other columns will
be considered as input columns. The target classes are depicted below in Table 2.
Sl.No Output
1. General-Hospital
2. Go Home
3. Intensive Care
TABLE 2: Target Classes
In this data set total numbers of variables are 9 and for data sub setting has been done using all
data rows. The total numbers of data rows are 88 and total weights for all rows are 88. From
analysis it has been noticed that number of rows with missing target or weight values are 0 and
rows with missing predictor values are 3 and were discarded because these variables had
missing values. Decision tree inducers are algorithms that automatically construct a decision tree
from a given dataset. Typically the goal is to find the optimal decision tree by minimizing the
generalization error. However, other target functions can be also defined, for instance, minimizing
the number of nodes or minimizing the average depth.

Target variable Discharge Decision
Number of predictor variables 8
Type of model Single tree
Maximum splitting levels 10
Type of analysis Classification
Splitting algorithm Gini
Category weights (priors) Data file distribution
Misclassification costs Equal (unitary)
Variable weights Equal
Minimum size node to split 10
Max. categories for continuous
predictors
200
Use surrogate splitters for missing
values
Yes
Always compute surrogate splitters No
Tree pruning and validation method Cross validation
Tree pruning criterion Minimum cost complexity
(0.00 S.E.)
Number of cross-validation folds 10
TABLE 3: Decision Tree Parameters
The Gini splitting algorithm is used for classification trees. Each split is chosen to maximize the
heterogeneity of the categories of the target variable in the child nodes. Cross Validation method
is used to evaluate the quality of the model. In this study, single decision tree model was built
with maximum splitting level is 10 and misclassification cost (Equal) are same for all
categories. Surrogate splitters are used for the classification of rows with missing values.
There are various top-down decision trees inducers such as ID3 [13], C4.5 [7], CART [8]. Some
consist of two conceptual phases: growing and pruning (C4.5 and CART). Other inducers perform
only the growing phase. In most of the cases, the discrete splitting functions are univariate.
Univariate means that an internal node is split according to the value of a single attribute.
Consequently, the inducer searches for the best attribute upon which to split. There are various
univariate criteria. These criteria can be characterized in different ways:
• According to the origin of the measure: information theory, dependence, and distance.
• According to the measure structure: impurity based criteria, normalized impurity based
criteria and Binary criteria.
Gini index is an impurity-based criterion that measures the divergences between the probability
distributions of the target attribute's values. In this work Gini index has been used and it is
defined as
(1)
Consequently the evaluation criterion for selecting the attribute ai is defined as:
(2)
The summary of variables section displays information about each variable that was present in
the input dataset. The first column shows the name of the variable, the second column shows
how the variable was used; the possibilities are Target, Predictor, Weight and Unused. The third
column shows whether the variable is categorical or continuous, the forth column shows how

many data rows had missing values on the variable, and the fifth column shows how many
categories (discrete values) the variable has, as shown in table 4.
TABLE 4: Summary of variables
In Table 4, Discharge Decision is a target variable and the values are to modeled and predicted
by other variables. There must be one and only one target variable in a model. The variables
such as InternalTemp, SurfaceTemp, OxygenSat, etc are predictor variables and the values of
this variable are used to predict the value of the target variable. In this study there are 8
predictor variables and 1 target variable are used to predict the model. All the variables except
comfort are categorical variables.
Maximum depth of the
tree
9
Total number of group
splits
16
No of terminal (leaf)
nodes of a tree
9
The minimum
validation relative
error occurs
with 9 nodes.
The relative error
value
1.1125
Standard error 0.1649
The tree will be
pruned
from 28 to 7
nodes.
TABLE 5: Model Size - Summary Report
Table 5 displays information about the maximum size tree that was built, and it shows summary
information about the parameters that were used to prune the tree.
Nodes
Val
Cost
Val
std err
RS
cost
Complexity
9 1.1125 0.1649 0.7200 0.000000
1 1.0000 0.0000 1.0000 0.009943
TABLE 6: Validation Statistics
No Variable Class Type Missing
rows
Categories
1 InternalTemp Predictor Categorical 0 3
2 SurfaceTemp Predictor Categorical 0 3
3 OxygenSat Predictor Categorical 0 2
4 BloodPress Predictor Categorical 0 3
5 SurfaceTempStab Predictor Categorical 0 2
6 IntTempStab Predictor Categorical 0 3
7 BloodPressStab Predictor Categorical 0 3
8 Comfort Predictor Continuous 3 4
9 Discharge
Decision
Target Categorical 0 3

Table 6 displays information about the size of the generated tree and statistics used to prune the
tree. There are five columns in the table:
Nodes – This is the number of terminal nodes in a particular pruned version of the tree. It will
range from 1 up to the maximum nodes in the largest tree that was generated. The maximum
number of nodes will be limited by the maximum depth of the tree and the minimum node size
allowed to be split on the Design property page for the model.
Val cost – This is the validation cost of the tree pruned to the reported number of nodes. It is the
error cost computed using either cross-validation or the random-row-holdback data. The
displayed cost value is the cost relative to the cost for a tree with one node.
The validation cost is the best measure of how well the tree will fit an independent dataset
different from the learning dataset.
Val std. err. – This is the standard error of the validation cost value.
RS cost – This is the re-substitution cost computed by running the learning dataset through the
tree. The displayed re-substitution cost is scaled relative to the cost for a tree with one node.
Since the tree is being evaluated by the same data that was used to build it, the re-substitution
cost does not give an honest estimate of the predictive accuracy of the tree for other data.
Complexity – This is a Cost Complexity measure that shows the relative tradeoff between the
number of terminal nodes and the misclassification cost.
Actual Misclassified
Category Count Weight Count Weight Percent Cost
General-
Hospital
63 63 1 1 1.587 0.0616
Go-Home 23 23 15 15 65.217 0.652
Intensive-
Care
2 2 2 2 100.00 1.000
Total 88 88 18 18 20.455 0.205
TABLE 7: Misclassification Table: Training Data
Table 7 shows the misclassifications for the training dataset and it describes the number of rows
for “General – Hospital” misclassified by the tree 1, for Go-Home is 15 and for Intensive-Care is 2.
This misclassification cost for Intensive-Care is high.
Actual Misclassified
Category Count Weight Count Weight Percent Cost
General-
Hospital
63 63 8 8 12.698 0.127
Go-Home 23 23 18 18 78.261 0.783
Intensive-
Care
2 2 2 2 100.00 1.000
Total 88 88 28 28 31.818 0.318
TABLE 8: Misclassification Table: Validation Data

Table 8 shows the misclassifications for the validation dataset and total misclassification percent
is 31.818%. A Confusion Matrix provides detailed information about how data rows are classified
by the model. The matrix has a row and column for each category of the target variable. The
categories shown in the first column are the actual categories of the target variable. The
categories shown across the top of the table are the predicted categories. The numbers in the
cells are the weights of the data rows with the actual category of the row and the predicted
category of the column. Table 9 & 10 shows confusion matrix for training data and for validation
data.
Predicated Category
Actual
Category
General-
Hospital
Go-Home
Intensive-
care
General-
Hospital
62 1 0
Go-Home 15 8 0
Intensive-
Care
2 0 0
TABLE 9: Confusion Matrix- Training Data
The numbers in the diagonal cells are the weights for the correctly classified cases where the
actual category matches the predicted category. The off-diagonal cells have misclassified row
weights.
Predicated Category
Actual Category
General-
Hospital
Go-Home
Intensive-
care
General-Hospital 58 8 0
Go-Home 18 5 0
Intensive-Care 1 1 0
TABLE 10: Confusion Matrix - Validation Data
A. Lift/Gain for Training Data
The lift and gain table is a useful tool for measuring the value of a predictive model. Lift and gain
values are especially useful when a model is being used to target (prioritize) marketing efforts.
The following 3 table’s shows Lift/Gain for the three different outputs such as General-Hospital,
Go-Home, and Intensive Care.
Lift/Gain for Discharge Decision = General-Hospital

Bin
Index
Class
% of
Bin
Cum %
Population
Cum
% of
Class
Gain
% of
Population
% of
class
Lift
1 88.89 10.23 12.70 1.24 10.23 12.70 1.24
2 66.67 20.45 22.22 1.09 10.23 9.52 0.93
3 77.78 30.68 33.33 1.09 10.23 11.11 1.09
4 88.89 40.91 46.03 1.13 10.23 12.70 1.24
5 77.78 51.14 57.14 1.12 10.23 11.11 1.09
6 77.78 61.36 68.25 1.11 10.23 11.11 1.09
7 100.00 71.59 82.54 1.15 10.23 14.29 1.40
8 66.67 81.82 92.06 1.13 10.23 9.52 0.93
9 44.44 92.05 98.41 1.07 10.23 6.35 0.62
10 14.29 100.00 100.00 1.00 7.95 1.59 0.20
TABLE 11: Lift and Gain for training data (General-Hospital)
Average gain = 1.112
Percent of cases with Discharge Decision = General-Hospital: 71.59%
Lift/Gain for Discharge Decision = Go-Home
Bin
Index
Class
% of
Bin
Cum %
Population
Cum
% of
Class
Gain
% of
Population
% of
class
Lift
1 88.89 10.23 34.78 3.40 10.23 34.78 3.40
2 0.00 20.45 34.78 1.70 10.23 0.00 0.00
3 33.33 30.68 47.83 1.56 10.23 12.04 1.28
4 11.11 40.91 52.17 1.28 10.23 4.35 0.43
5 22.22 51.14 60.87 1.19 10.23 8.70 0.85
6 22.22 61.36 69.57 1.13 10.23 8.70 0.85
7 22.22 71.59 78.26 1.09 10.23 8.70 0.85
8 22.22 81.82 86.96 1.06 10.23 8.70 0.85
9 11.11 92.05 91.30 0.99 10.23 4.35 0.43
10 28.57 100.00 100.00 1.00 7.95 8.70 1.09
TABLE 12: Lift and Gain for training data (Go-Home)
Percent of cases with Discharge Decision = Go-Home: 26.14%
Lift/Gain for Discharge Decision = Intensive-Care
Bin
Index
Class
% of
Bin
Cum %
Population
Cum
% of
Class
Gain
% of
Population
% of
class
Lift
1 2.27 10.00 10.00 1.00 10.00 10.00 1.00
2 2.27 20.00 20.00 1.00 10.00 10.00 1.00
3 2.27 30.00 30.00 1.00 10.00 10.00 1.00
4 2.27 40.00 40.00 1.00 10.00 10.00 1.00
5 2.27 50.00 50.00 1.00 10.00 10.00 1.00
6 2.27 60.00 60.00 1.00 10.00 10.00 1.00
7 2.27 70.00 70.00 1.00 10.00 10.00 1.00
8 2.27 80.00 80.00 1.00 10.00 10.00 1.00
9 2.27 90.00 90.00 1.00 10.00 10.00 1.00
10 2.27 100.00 100.00 1.00 10.00 10.00 1.00

TABLE 13: Lift and Gain for training data (Intensive -Care)
Percent of cases with Discharge Decision = Intensive-Care: 2.27%
Lift and Gain for validation data
Lift/Gain for Discharge Decision = General-Hospital
Bin
Index
Class
% of
Bin
Cum %
Population
Cum
% of
Class
Gain
% of
Population
% of
class
Lift
1 77.78 10.23 11.11 1.09 10.23 11.11 1.09
2 55.56 20.45 19.05 0.93 10.23 7.94 0.78
3 77.78 30.68 30.16 0.98 10.23 11.11 1.09
4 77.78 40.91 41.27 1.01 10.23 11.11 1.09
5 88.89 51.14 53.97 1.06 10.23 12.70 1.24
6 88.89 61.36 66.67 1.09 10.23 12.70 1.24
7 100.00 71.59 80.95 1.13 10.23 14.29 1.40
8 44.44 81.82 87.30 1.07 10.23 6.35 0.62
9 44.44 92.05 93.65 1.02 10.23 6.35 0.62
10 57.14 100.00 100.00 1.00 7.95 6.35 0.80
TABLE 14: Lift and Gain for Validation data (General Hospital)
Percent of cases with Discharge Decision = General-Hospital: 71.59%
Lift/Gain for Discharge Decision = Go-Home
Bin
Index
Class
% of
Bin
Cum %
Population
Cum
% of
Class
Gain
% of
Population
% of
class
Lift
1 22.22 10.23 8.70 0.85 10.23 8.70 0.85
2 44.44 20.45 26.09 1.28 10.23 17.39 1.70
3 44.44 30.68 43.48 1.42 10.23 17.39 1.70
4 33.33 40.91 56.52 1.38 10.23 13.04 1.28
5 22.22 51.14 65.22 1.28 10.23 8.70 0.85
6 22.22 61.36 73.91 1.20 10.23 8 .70 0.85
7 11.11 71.59 78.26 1.09 10.23 4.35 0.43
8 22.22 81.82 86.96 1.06 10.23 8.70 0.85
9 22.22 92.05 95.65 1.04 10.23 8.70 0.85
10 14.29 100.00 100.00 1.00 7.95 4.35 0.55
TABLE 15: Lift and Gain for Validation data (Go-Home)
Percent of cases with Discharge Decision = Go-Home: 26.14%
Lift/Gain for Discharge Decision = Intensive-Care

Bin
Index
Class
% of
Bin
Cum %
Population
Cum
% of
Class
Gain
% of
Population
% of
class
Lift
1 2.27 10.00 10.00 1.00 10.00 10.00 1.00
2 2.27 20.00 20.00 1.00 10.00 10.00 1.00
3 2.27 30.00 30.00 1.00 10.00 10.00 1.00
4 2.27 40.00 40.00 1.00 10.00 10.00 1.00
5 2.27 50.00 50.00 1.00 10.00 10.00 1.00
6 2.27 60.00 60.00 1.00 10.00 10.00 1.00
7 2.27 70.00 70.00 1.00 10.00 10.00 1.00
8 2.27 80.00 80.00 1.00 10.00 10.00 1.00
9 2.27 90.00 90.00 1.00 10.00 10.00 1.00
10 2.27 100.00 100.00 1.00 10.00 10.00 1.00
TABLE 16: Lift and Gain for Validation data (Intensive Care)
Percent of cases with Discharge Decision = Intensive-Care: 2.27%
Table 17 shows the relative importance of each predictor variables. The calculation is performed
using sensitive analysis where the values of each variables are randomized and the effect on the
quality of the model is measured.
Sl.No Variable Importance
1 Comfort 100.00
2 BloodPressStab 51.553
3 InternalTemp 34.403
4 SurfaceTemp 25.296
5 IntTempStab 20.605
6 BloodPress 20.045
TABLE 17: Over all Importance of Variables
The results of this experimental research through the above critical analysis using this decision
tree based method shows more accurate data prediction and helping the clinical experts in taking
decisions such as General Hospital, Go-Home and Intensive-care. More accuracy will be
guaranteed for large data sets.
5. CONCLUSION
This paper has presented a decision tree classifier to determine the patient’s post-operative
recovery decisions such as General Hospital, Go-Home and Intensive-care. Figure 2 depicts the
decision tree model for the prediction of patient’s post-operative recovery area.

FIGURE 2: Decision tree for the prediction of post-operative recovery area
In figure 2, each node represents set of records from the original data set. Nodes that have child
nodes such as 1, 2, 3, 4, 25, 7, 26, 9 are called interior nodes. Nodes 5, 24,6,27,8,28,29,18,19
are called leaf nodes. Node 1 is root node. The value N in each node shows number of rows
that fall on that category. The target variable is Discharge Decision. Misclassification percentage
in each node shows the percentage of the rows in this node that had target variable categories
different from the category that was assigned to the node. In other words, it is the percentage of
rows that were misclassified. For instance, from the tree it is clearly understood the value of
predictor variable comfort is <= 6 the decision is Go-Home. If Comfort is >6 additional split is
required to classify. The total time consumed for analysis is 33ms.
The Decision tree analysis discussed in this paper highlights the patient sub-groups and critical
values in variables assessed. Importantly the results are visually informative and often have clear
clinical interpretation about risk factors faced by these subgroups. The results shows 71.59% of
cases with General-Hospital 26.14% of cases with Go-Home and 2.27% cases with Intensive-
Care.
6. REFERENCES
1. Banning, M. (2007). “A review of clinical decision making models and current research,”
Journal of clinical nursing, [online] available
at:http://www.blackwellsynergy.com/doi/pdf/10.1111/j.1365-2702.2006.01791.x
2. T. af Klercker (1996): “Effect of Pruning of a Decision-Tree for the Ear, Nose and Throat
Realm in Primary Health Care Based on Case-Notes”. Journal of Medical Systems, 20(4):
215-226.

3. M. Zorman, V.Podgorelec, P.Kokol, M.Peterson, J. Lane (2000). “Decision tree's induction
strategies evaluated on a hard real world problem.” In: 13th IEEE symposium on computer-
based medical systems 22-24 June 2000, Houston, Texas, USA: proceedings, Los Alamitos,
IEEE Computer Society 19-24.
4. Shanthi D, Sahoo G, Saravanan N. (2008). “Input Feature Selection using Hybrid Neuro-
Genetic Approach in the diagnosis of Stroke.” International Journal of Computer Science and
Network Security, ISSN: 1738-7906, 8(12):99-107.
5. F.S,Khan, R,M,Anwer, of Torgersson, and G. Falkman (2009), “Data Mining in Oral Medicine
Using Decision Trees”, International Journal of Biological and Medical Sciences 4:3.
6. Shanthi D, Sahoo G, Saravanan N. (2009), ”Comparison of Neural Network Training
Algorithms for the prediction of the patient’s post-operative recovery area”, Journal of
Convergence Information Technology, ISSN: 1975-9320,4(1):24-32.
7. Quinlan JR (1993) C4.5: “programs for machine learning”. California: Morgan Kaufmann
Publishers.
8. Breiman L, Friedman JH, Olshen RA, et al., (1984). “Classification and regression trees.”
Belmont Calif: Wadsworth International Group.
9. Gehrke, J., Ramakrishnan, R., & Ganti, V. (1998). “Rain-Forest A framework for fast decision
tree construction of large datasets”. Proceedings of the 24
th
International Conference on Very
Large Data Bases, pp. 416–427.
10. Shafer, J., Agrawal, R., & Mehta, M. (1996). “SPRINT: A scalable parallel classifier for data
mining.” Proceedings of the 22
th
International Conference on Very Large Data Bases, pp.
544–555.
11. “UCI Machine Learning Repository” [online] available at:.
(http://www.ics.uci.edu/~mlearn/MLSummary.html). The data was originally created by
Sharon Summers (School of Nursing, University of Kansas) and Linda Woolery (School of
Nursing, University of Missouri) and donated by Jerzy W. Grzymala-Busse.
12. Abdel-badeeh M.Salem, Abeer M.Mahmoud (2002). “A hybrid genetic algorithm –Decision
tree classifier,” IJICIS, 2(2)
13. Quinlan, J. R. (1986). “Induction of decision trees”. Machine Learning, 1, 81–106.
14. DTREG version 9.1 “Decision Tree Software for Predictive Modeling and Forecasting”.
15. Fayyad, U., Piatetsky-Shaprio, G., Smyth, P. & Uthurusamy, R. (1996). (eds.),” Advances in
Knowledge Discovery and Data Mining”, MIT Press, Cambridge, MA
16. Hillier, F. & Lieberman, G. (2001). “Introduction to Operations Research,” McGrawHill,
Boston.
17. Abbass, H.A., Towsey M., &Finn G. (2001). C-Net: “A Method for Generating Non-
deterministic and Dynamic Multivariate Decision Trees.” Knowledge and Information
Systems: An International Journal, Springer-Verlag, 5(2).

Decision Tree Classifiers to determine the patient’s Post-operative Recovery Decision

More Related Content

What's hot (20)

Similar to Decision Tree Classifiers to determine the patient’s Post-operative Recovery Decision (20)

More from Waqas Tariq (20)

Recently uploaded (20)

Decision Tree Classifiers to determine the patient’s Post-operative Recovery Decision