The bankruptcy-prediction problem can be viewed as a problem of classification

The bankruptcy-prediction problem can be viewed as a problem of classification

The decision tree is essential in an organization to specific departments; a marketing manager may wish to have a set of consumers who are likely to purchase his/her product. He/she can use the decision tree to acquire enough and large numbers of an audience while saving the marketing budget. Lenders’ managers are another department in an institution that can use the decision tree to categorize consumers into groups of non-potential and potential or risky or safe loan applications as a problem classification. The process of classification is a two-step, prediction, and learning step. In the learning stage, the model is created, relying on given training information. In the prediction stage, the model is used to forecast the reply of the provided information. Therefore, decision tree algorithms are one of the famous and most accessible classifications to interpret and understand. It can be employed for both regression and classification kind of difficulties.

Decision Tree Algorithm

A decision tree can be described as a flowchart like configuration where the most internal node symbolizes the attribute, the branch signifies a decision rule, and each leaf stands for the outcome. The topmost node expresses the root node in a decision tree. The decision tree learns to set differences through the attribute value basis. This structure looks like a flowchart to assist you in making a decision. It is a flowchart-like diagram that envisages and easily caricaturist the level of human thinking.

The decision tree distributes the logic of making a decision, which is not available in some type of algorithms like Neural Network. It consumes little time during training as associated with the algorithm of neural networks. The decision trees’ time complexity is a function of the number and record and amount of attributes in the given information. The decision tree is a non-parametric or distribution-free technique, which does not rely on probability distribution conventions. With the decision tree, even the high dimensional information can be handled with 100% accuracy.

How the Decision Tree algorithm functions

1. Choose the best attribute by the use of Attribute Selection Measures to divide the records.
2. Attribute a decision node and break the data set into smaller subsets.
3. Repeat this process recursively to commence tree building for every child until condition equals:

• All the tuples belong to the same value of the attribute.
• No more remaining attributes.
• Ensure no more occasions.

Selection Measures of Attributes

This refers to an experimental for choosing the splitting technique that partition information into the most elegant manner possible. It assists us in determining the tuples’ breakpoint on a particular node. It also helps in providing rank to each attribute by explaining a specific set of data. The most common measures of selection are; Gain Ratio, Information Gain, and Gini Index.

Information Gain

Shannon invented this concept by use of entropy, which is responsible for measuring the impurity of the data set. Information Gain is the entropy decrease. Decision Tree algorithm makes use of information gain.

Where

Info(D) is the mean number of data required to identify the class label of a tuple in D

|Dj|/|D| acts as the jth partition weight

InfoA(D) is the anticipated data needed to categorize a tuple from D based on partitioning by A

Attribute A with the maximum data gain, Gain(A), is selected as the splitting attribute at node N().

Gain Ratio

Information gain is biased as it takes the attribute with many outcomes. Gain ratio handles this issue by normalizing data obtained using split info.

Where,

|Dj|/|D| acts as jth partition’s weight

V is the number of discrete value in attribute A

The gain ratio can be concluded as

The attribute with the highest gain ratio is selected as the splitting attribute

Gini Index

It is another technique in decision tree algorithm used to build split points

Where Pi is the probability that tuple in D belongs to class Ci.

It puts more consideration on the binary split of every attribute. If binary divide on a quality A partitions information D into D1 and D2, the Gini index of D is

Neural Network

They are used to handle most difficulties associated with artificial intelligence; they frequently outperform ancient machine education models because they have the benefit of non-linearity, customizability, and variable interactions.

Neural Network

The process of neural network creation commences with the perceptron. The perceptron obtains inputs, increase them by several weights, and then forwards them into an activation function (such as ash, relu, identity, and logistics) to give an output. A neural network is formed by the addition of the perceptrons layer together, referred to as a model of a multi-layer perceptron. The neural network consists of three layers- the hidden, the output, and the input. The segments that are in between are known as the hidden layer where the intermediate computation occurs. A neural network algorithm can be used for both regression and classification difficulties.

Problem statement

To create a dashboard, consumption of diabetes dataset which contains 768 observations and nine variables will be applied as follows;

1. Glucose-concentration of glucose in plasma
2. Triceps- thickness of skinfold (mm)
3. Age – age (years)
4. Pregnancies – count of pregnancies time
5. Diastolic – blood pressure (mm Hg)
6. Diabetes – “I” for the presence and “O” for absent
7. Insulin – insulin serum per hour (mu U/ml)
8. Dpf – diabetes pedigree function
9. Bmi – basal metabolic rate

Steps

1. Loading the necessary modules and libraries
2. Reading information and performing necessary information check

The beginning line of the code indicates information as a pandas data frame, while the second line is responsible for prints the shape – 769 observations of 9 variables. The third line shows transposed synopsis information of the variables.

Output

3. Formulating arrays for the response variables and features

The first code line produces on an object of the target variable known as ‘target-column.’ The second code line shows us all features, not including the target variable ‘unemploy,’ while the third code line normalized the predictor.

The fourth code line indicates the review of the normalized information.

Output

4. Creating the test datasets and training

The first pair of code lines produces arrays of the independent (X) and dependent (Y) variables, respectively. The third line divides the information into test dataset and training, and the fourth code line prints the shape of the test data and training.

Output

5. Evaluating, Predicting, and Creating the model of Neural Network.

The third code line fits the training information model, while the fourth and fifth line uses the trained model to produce predictions on the test dataset and training.

Once the predictions are produced, which is done in the first code line below, the second and third code line prints the perplexity matrix, and the uncertainty reports outcomes on the training information.

Output

The above output indicates the performance of the model on training information. The F1 score and the accuracy score is around 0.77 and 0.78, respectively.

The next stage is to analyze the model performance on the test data performed with the code line below.

Output

The above result indicates the performance test data model, the F1 scores and accuracy both around 0.75

References

https://www.datacamp.com/community/tutorials/decision-tree-classification-python

: https://www.pluralsight.com/guides/machine-learning-neural-networks-scikit-learn

Fallahpour, S., Lakvan, E.N., and Zadeh, M.H., 2017. Using an ensemble classifier based on sequential floating forward selection for financial distress prediction problem. Journal of Retailing and Consumer Services, 34, pp.159-167.

García, V., Marqués, A.I., Sánchez, J.S. and Ochoa-Domínguez, H.J., 2019. Dissimilarity-Based Linear Models for Corporate Bankruptcy Prediction. Computational Economics, 53(3), pp.1019-1031.

Ouenniche, J., Bouslah, K., Cabello, J.M., and Ruiz, F., 2018. A new classifier based on the reference point method with application in bankruptcy prediction. Journal of the Operational Research Society, 69(10), pp.1653-1660.