Decision Tree Method Using for Fetal State Classification from Cardiotography Data

. The motive of the investigation is analyzing the categorization of fetal state code from the Cardiographic data set based on decision tree method. Cardiotocography is one of the important tools for monitoring heart rate, and this technique is widely used worldwide. Cardiotocography is applied for diagnosing pregnancy and checking fetal heart rate state condition until before delivery. This classi(cid:28)cation is necessary to predict fetal heart rate situation which is belonging. In this paper, we are us-ing three input attributes of training data set quoted by LB, AC, and FM to categorize as normal, suspect or pathological where NSPF variable is used as response variable. After drawing necessary analyzing into three variables we get the 19 nodes of classi(cid:28)cation tree and also we have measured every single node according to statistic, criterion, weights and values. The Cardiotocography Dataset applied in this study are received from UCI Machine Learning Repository. The dataset contains 2126 observation instances with 22 attributes. In this experiment, the highest accuracy is 98.7%. Overall, the experimental results proved the viability of Classi-(cid:28)cation and Regression Trees and its potential for further predictions


Introduction
rate is the most exploited information and can't be measured directly during pregnancy because of fetus innate inaccessibility. Necessarily, clinicians depend on indirect symptoms and information to learn about fetal condition [2]. analysis methods have been proposed [5]. As example we will notice that some automated software helps clinicians by providing signal-processing facilities to determine and measure some employed parameters like vibrancy, mutability, stimulation of fetal heart rate, frequency of UC etc. Although, the role of these computerized classication processes have only partially reduced the inter-and intra-observer variations [6].
For the fetus, the accepted parameters of Baseline fetal heart rate are started in between 110 and 160 where variance is greater than 5 [7]. Clinicians nd information like continual heart rate indication and fetal heart rate variations in response of CTG. It is one kind of indirect screening test on fetal acid-base balance and is regarded as an indicator of fetal welfare [8].
Since its launch in the 1970s, CTG is widely used in daily practice. Clinicians expected that fetal heart rate monitoring would work as a screening test for acute asphyxia which is enough to cause of neurologic damage or death of newborn [9,10].
While it's practically, the eectiveness and success of CTG monitoring has some inconsistencies, specically in low-risk pregnancies. If fetal pain is examined inaccurately, it may be mentioned to as needless treatments. On the other hand, an inappropriate analysis of fetal well-being may be refused necessary treatments [11].
The classication is used to measure performance evaluation, actually it's not enough to make a conclusion for such an important situation as the eld of medical diagnosis. Because of that, it is proposed another kind of implement measurement tools as like the decision tree. In this paper, our vision is to analyze CTG data set based on fetal classication code by using decision tree method.
Decision tree (DT) is a process to constitute an alignment model. Like the name it suggests, a tree-shaped form is built using inductive argument. Decision tree has turned to be a very fruitful appliance for performing medical prophecy models [12].

Training data set
In this study, mentioned cardiotocography data set has taken from The Data Mining Repository of University of California Irvine (UCI) [14]. This training data set has consisted of 2126 samples fetal cardiotocograms which belongs to dierent classes with 21 attributes and two class labels.
In this training, we used fetal state class code as object point attribute in exchange to fetal heart rate sampling group and each group has illustrated into three dierent types as normal, suspect and pathological.
The Attribute information is given Tab. 1.

Proposed methodology
Classication and regression Trees is called CART or Tree. CART is one of the renowned supervised learning algorithms. It is a choice support tool that is employed as a tree-like pattern of selections and their possible consequences, resource values, and utility. This is a way of displaying algorithms that are only represented as a binary tree.
Decisions trees are usually utilized in research, to assist determine the technique possible to achieve a goal, however also are a well-liked tool in machine learning algorithm.
Classication and regression trees (CART) regarded as a decision tree technic employed for classication motive using the diachronic data.
As a machine learning method, it is introduced in 80s by and is considered to ascertain number of classes [15]. To form decision trees, CART requires learning sample. Decision trees are a set of questions, which is divided the learning example into short and unconsidered parts. In order to acquire the best divide, the query that divides the data into 2 kindred portions, CART system explores of all probable variables and values [16].
CART is a binary decision tree approach and shows the direction for another algorithms like Random Forest, Bagged Trees and Boosted trees, where every root node explains a single input variable and a divide point on that variable. The leaf nodes comprise the response variable which is used to establish predictions [17].
CART uses Gini index to pick out the attribute which has maximum info. A data "A" with n categories has Gini Index that's outlined as  [19]. Here C4.5 can be taken as an instance of that system which instructs decision-tree classiers method [20].
For getting new measurement here we utilize c4.5 that is also known as Gain ratio. It can be described by below formula, When a decision tree has been formed, in some branches exceptions in the training data will ap- techniques [21,22]. In a decision tree, data classication has two-phase operation where training phase is the rst one, and another one is classication phase. At rst phase, a training data is assisted for composition of the tree and trees rules are ascertained in accordance with this training data [23].
In this thesis for sampling training data set we follow a conceptual process.  The tree (in Fig. 2) generally grows upwards and downwards. If we look at the graph, roots are at the top and leaps are bottom. Basically most important roots will be stand at top position of prediction models. In this classication most important variable is AC, that's why its position is above and each of the three variables helps to classify 1, 2 and 3 which 68 c 2020 Journal of Advanced Engineering and Computation (JAEC) VOLUME: 4 | ISSUE: 1 | 2020 | March are represented normal, suspect and pathological. If AC is <0.001, it will go to left side and if more than >0.001, it will go on right side. Similarly if LB is <136, will go to left side and more than >136 will go on right side. Final variables FM is <0.1 which will go terminal node and this is the decision tree. Our total validate data set has 408 observations. For predict into the validation data set based on the tree we will use this 408 observations data. We had trained this data for getting our experimental results that how much probability to get normal, suspect and pathological. The results of this data prediction are given below Here, we just represented only last 10 data results out of 408 observations. If we look at predict data, we can see there are three types of categories, 1st one is showing probability of patients will be normal, similarly 2nd one is suspected and last one is pathological. We can also use this validate data set for prediction without probability. We have used rpart packages for validate data based on decision tree. Graphically it will be Fig. 3. Figure 3 shows that predict sample accordingly three variables. If we look the AC then we will see 934 patients are normal, 33 is suspect and 21is pathological. Similar to others two features. Here we used extra 1 for seeing the result that how its show the graphically and same to way we used the extra value 2, 3 and 4 so it will show changes. Again if we used the extra two then it will be show dierences. The dierent graph is given Fig. 4.

Discussions
In this train data set we get classication error is 0.19798.And out of 408 validate data observation we get the classication error is 0.2107.   For this graph we choose randomly data .Every features we sampling 50 data .In this graph if we analysis this three features with response variable it's clearly represent the dierences of them. LB features data is represented FHR baseline (beats per minute). Mentioned Fig. 4 shows the LB vs NSP (class code) variables gured. Similarly reaming two features AC and FM dened the FHR accelerations per second and fetal movements per second.
The cardiotocography enrolls the heart rate mutability, uterine compressions and fetal activities simultaneously during pregnancy. It gives a gynecologist knowledge about the cardiovascular system of fetal and the fetal activities [24]. Women who have bleeding, high blood pressure, premature labor or diabetes, CTG is specially recommended for them. An obstetrician can understand the CTG report through some key parameters. + Fetal heart rate Baseline (BL) 110-150 beats every 60 seconds (bpm) are in the range of physiological values. If the BL>150 bmp, that means the tachycardia and if the BL<110 bmp, that means the bradycardia. + Acceleration (ACC) for at least 15 seconds rises BL upper than 15 bpm. In every 15 minutes the acceleration should happen at least double. Dur-ing the night time acceleration is regarded as pathological condition and at the same time it could be responded to fetal activity. + Deceleration (DCL) -for at least 15 seconds decrease BL higher than 15 bpm. Along with constrictions it may imply birth asphyxia.

Conclusion
Machine learning is one of the best choice for medical diagnosis and prediction. In this study, we are applied several machine learning techniques and R programming simulation software to accomplish this approach, classication of cardiotocography. After applying suitable machine learning technique when data set is trained, the prediction is acquired certain feature. This research methodology may help to related people who are working about embryology.
This study focused only on fetal conditions where the health status of both the mother and the fetus can be considered and studied in future. The nobility of this research is to trace the fetal status during pregnancy period especially for the complicated cases. "This is an Open Access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium provided the original work is properly cited (CC BY 4.0)."