Assess the risk of company bankruptcy through machine learning

1. Application type

This is a classification project since the variable to predict is binary (bankruptcy or non-bankruptcy).

The goal is to model the probability of a business going bankrupt based on various features.

2. Data set

The data set contains information to create our model. We need to configure three things:

• Data source.
• Variables.
• Instances.

Data source

The data file used for this example is bankruptcy-prevention.csv, which contains 7 features about 250 companies.

Variables

The data set includes the following variables:

• industrial_risk: 0=low risk, 0.5=medium risk, 1=high risk.
• management_risk: 0=low risk, 0.5=medium risk, 1=high risk.
• financial_flexibility: 0=low flexibility, 0.5=medium flexibility, 1=high flexibility.
• credibility: 0=low credibility, 0.5=medium credibility, 1=high credibility.
• competitiveness: 0=low competitiveness, 0.5=medium competitiveness, 1=high competitiveness.
• operating_risk: 0=low risk, 0.5=medium risk, 1=high risk.
• class: bankruptcy, non-bankruptcy (target variable).

Instances

On the other hand, the instances are divided randomly into training, selection, and testing subsets, containing 60%, 20%, and 20% of the instances, respectively.

Distributions

Our target variable is class. We can calculate the data distributions and plot a pie chart with the percentage of instances for each class.

As we can see, the target variable is well-balanced. Indeed, 42.8% are positive samples and 57.8% are negative samples.

Input-target correlations

The correlations between inputs and targets might indicate which factors have the most significant influence on going into bankruptcy.

In this example, the variables that correlate the most with the target variable negatively correlate. Competitiveness is the variable with the highest correlation.

3. Neural network

The next step is to set the neural network parameters. For classification problems, it is composed of:

• Scaling layer.
• Dense layer.

Scaling layer

The mean and standard deviation scaling method has been set for the scaling layer.

Dense layer

We set up a single dense layer with one neuron, utilizing the logistic activation function. Since the target variable is binary, this layer has six inputs and only one output.

Neural network graph

The neural network for this example can be represented with the following diagram:

4. Training strategy

The fourth step is to set the training strategy, defining what the neural network will learn. A general training strategy for classification is composed of two terms:

• A loss index.
• An optimization algorithm.

Loss index

The loss index chosen for this problem is the normalized squared error between the outputs from the neural network and the targets in the data set with L1 regularization.

Optimization algorithm

The selected optimization algorithm is the Quasi-Newton method.

Training

The following chart shows how training and selection errors develop with the epochs during training.

The final values are training error = 0.035 NSE and selection error = 0.028 NSE.

5. Testing analysis

The objective of the testing analysis is to validate the generalization performance of the trained neural network.

To validate a classification technique, we need to compare the values provided by this technique to the observed values.

ROC curve

We can use the ROC curve as it is the standard testing method for binary classification projects.

The AUC value for this example is 1.

Confusion matrix

The following table contains the elements of the confusion matrix.

This matrix contains the variable class’s true positives, false positives, false negatives, and true negatives.

The total number of testing samples is 50, all correctly classified.

Cumulative gain

We can also perform a cumulative gain analysis, a visual tool that compares a predictive model against random selection.

The chart includes three lines.

• The baseline represents what would be obtained by chance.
• The positive cumulative gain (y-axis) shows the percentage of positives found for each portion of the population (x-axis).
• The negative cumulative gain shows the percentage of negatives found in the same way.

7. Model deployment

In the model deployment phase, we use the neural network to predict the risk of a company.

References

• The data for this problem has been taken from the Machine Learning UCI Repository.

Related posts