Contents<\/h3>\n\n- Application type<\/a>.<\/li>\n
- Data set<\/a>.<\/li>\n
- Neural network<\/a>.<\/li>\n
- Training strategy<\/a>.<\/li>\n
- Model selection<\/a>.<\/li>\n
- Testing analysis<\/a>.<\/li>\n
- Model deployment<\/a>.<\/li>\n
- Tutorial video<\/a>.<\/li>\n<\/ol>\n
This example was solved with Neural Designer<\/a>. You can use the free trial<\/a> to follow this example step by step.<\/p>\n<\/section>\n\n1. Application type<\/h2>\n
This is a classification<\/a> project because the predictor variable is binary (fraudulent or not).<\/p>\nThe goal is to develop a model that estimates the likelihood of a transaction being fraudulent.<\/p>\n<\/section>\n\n2. Data set<\/h2>\n
The dataset contains the information needed to create our model. We need to configure three things:<\/p>\n
\n- Data source.<\/li>\n
- Variables.<\/li>\n
- Instances.<\/li>\n<\/ul>\n
The data file used for this example is creditcard-fraud.csv<\/a>, which contains 11 features for about 3,075 payments.<\/p>\nVariables<\/h3>\n
The data set includes the following variables<\/a>:<\/p>\n<\/section>\nMerchant Behavior<\/strong><\/h4>\n\n- \n
Merchant ID<\/strong> (merchant_id<\/code>): Unique identifier of the merchant where the transaction occurred.<\/p>\n<\/li>\n<\/ul>\nTransaction Data<\/strong><\/h4>\n\n- Transaction Amount<\/strong> (
transaction_amount<\/code>): Monetary value of the individual transaction.<\/li>\n- Average Daily Transaction Amount<\/strong> (
avg_amount_day<\/code>): Mean value of transactions made with the card during a given day.<\/li>\n- Foreign Transaction<\/strong> (
foreign_transaction<\/code>): Indicates whether the transaction was made outside the cardholder\u2019s country (Yes\/No).<\/li>\n- High-Risk Country<\/strong> (
high_risk_country<\/code>): Flags if the transaction originated from a country considered high-risk for fraud (Yes\/No).<\/li>\n- Declined Transaction<\/strong> (
is_declined<\/code>): Specifies if the credit card transaction was declined (Yes\/No).<\/li>\n- Daily Declines Count<\/strong> (
number_declines_day<\/code>): Total number of declined transactions in a single day.<\/li>\n<\/ul>\nFraud Indicators & Historical Data<\/strong><\/h4>\n\n- Daily Average Chargeback Amount<\/strong> (
daily_chbk_avg_amt<\/code>): Average monetary value of chargebacks recorded per day.<\/li>\n- 6-Month Average Chargeback Amount<\/strong> (
6m_avg_chbk_amt<\/code>): Mean chargeback amount accumulated over the last six months.<\/li>\n- 6-Month Chargeback Frequency<\/strong> (
6m_chbk_freq<\/code>): Number of chargebacks recorded in the last six months.<\/li>\n<\/ul>\nTarget variable<\/strong><\/h4>\n\n- Fraudulent Transaction<\/strong>\u00a0(
is_fraudulent<\/code>): Class label indicating whether the transaction was fraudulent or not.<\/li>\n<\/ul>\n\nInstances<\/h3>\n
On the other hand, the instances<\/a> are divided randomly into training, selection, and testing subsets, containing 60%, 20%, and 20% of the instances, respectively.<\/p>\nVariables distributions<\/h3>\n
Our target variable is is_fraudulent<\/b>. We can calculate the data distributions<\/a> and plot a pie chart with the percentage of instances for each class.<\/p>\n![](\"https:\/\/www.neuraldesigner.com\/images\/credit_card_piechart.webp\")
<\/p>\n
As we can see, the target variable is unbalanced, with many payments being non-fraudulent (approximately 85%), while only 15% are fraudulent. We could say that approximately 1 out of 6 payments is fraudulent.<\/p>\n
Inputs-targets correlations<\/h3>\n
The input-target correlations might indicate which factors have the most significant influence on a fraudulent transaction.<\/p>\n
![](\"https:\/\/www.neuraldesigner.com\/images\/credit_card_correlations.webp\")
<\/p>\n
In this example, all variables have a positive correlation except for is_declined<\/i>. Moreover, the variable high_risk_country<\/i> has the highest correlation with the target variable.<\/p>\n<\/section>\n\n3. Neural network<\/h2>\n
The next step is to set the neural network<\/a> parameters. For classification problems, it is composed of:<\/p>\n\n- Scaling layer.<\/li>\n
- Perceptron layers.<\/li>\n
- Probabilistic layer.<\/li>\n<\/ul>\n
Scaling layer<\/h3>\n
We have set the mean and standard deviation scaling method<\/a> for the scaling layer<\/a>,<\/p>\nDense layers<\/h3>\n
We set\u00a0up one\u00a0perceptron layer<\/a> with 3 neurons, using<\/span>\u00a0the logistic activation function<\/a>. This layer has nine inputs, and since the target variable is binary, it has only one output.<\/p>\nThe neural network for this example can be represented with the following diagram:<\/p>\n
![](\"https:\/\/www.neuraldesigner.com\/images\/credit_card_nn.webp\")
<\/p>\n<\/section>\n\n4. Training strategy<\/h2>\n
The fourth step is to set the training strategy<\/a>, defining\u00a0what the neural network will learn. A general training strategy for classification is composed of two terms:<\/p>\n\n- A loss index.<\/li>\n
- An optimization algorithm.<\/li>\n<\/ul>\n
Loss index<\/h3>\n
The loss index<\/a> chosen for this problem is the normalized squared error<\/a> between the neural network’s outputs and the targets in the data set with L1 regularization.<\/p>\nThe selected optimization algorithm<\/a> is the Quasi-Newton method<\/a>. The selected optimization algorithm<\/a> is the Quasi-Newton method<\/a>.<\/p>\nThe following chart illustrates how training and selection errors evolve over the course of training epochs. The final values are\u00a0a training error\u00a0<\/strong>of\u00a00.052\u00a0<\/strong>and a\u00a0selection error\u00a0<\/strong>of 0.103, both measured in<\/strong><\/span>\u00a0NSE<\/b>.<\/p>\n![](\"https:\/\/www.neuraldesigner.com\/images\/creditcard_errorchart.webp\")
<\/p>\n<\/section>\n\n5. Model selection<\/h2>\n
The objective of model selection<\/a> is to find the network architecture with the best generalization properties, which means the one that minimizes the error on the selected instances<\/a> of the data set.<\/p>\nMore specifically, we want to find a neural network with a selection error smaller than 0.103 NSE<\/b>, which is the value we have achieved.<\/p>\n
Order selection<\/a> algorithms train several network architectures with a different number of neurons and select the one with the smallest selection error.<\/p>\nThe incremental order<\/a> method starts with a few neurons and increases the complexity at each iteration. The following chart shows the training error (blue) and the selection error (orange) as a function of the number of neurons.<\/p>\n![](\"https:\/\/www.neuraldesigner.com\/images\/creditcard_selectionerror.webp\")
<\/p>\n
The selection errors achieved are similar for any number of variables; however, the smallest is 0.1007<\/b> for an optimal number of neurons of 4.<\/p>\n
![](\"https:\/\/www.neuraldesigner.com\/images\/credit_card_final_nn.webp\")
<\/p>\n
The graph above represents the architecture of the final neural network.<\/p>\n<\/section>\n\n6. Testing analysis<\/h2>\n
The objective of the testing analysis<\/a> 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.<\/p>\nROC curve<\/h3>\n
We can use the ROC curve<\/a>\u00a0as it is the standard testing method for binary classification projects.<\/p>\n![](\"https:\/\/www.neuraldesigner.com\/images\/creditcard_ROC.webp\")
<\/p>\n
The AUC value for this example is 0.998.<\/p>\n
Confusion matrix<\/h3>\n
The following table contains the elements of the confusion matrix<\/a>.<\/p>\nThis matrix contains the true positives, false positives, false negatives, and true negatives for the variable is_fraudulent<\/i>.<\/p>\n
This example was solved with Neural Designer<\/a>. You can use the free trial<\/a> to follow this example step by step.<\/p>\n<\/section>\n This is a classification<\/a> project because the predictor variable is binary (fraudulent or not).<\/p>\n The goal is to develop a model that estimates the likelihood of a transaction being fraudulent.<\/p>\n<\/section>\n The dataset contains the information needed to create our model. We need to configure three things:<\/p>\n The data file used for this example is creditcard-fraud.csv<\/a>, which contains 11 features for about 3,075 payments.<\/p>\n The data set includes the following variables<\/a>:<\/p>\n<\/section>\n Merchant ID<\/strong> ( On the other hand, the instances<\/a> are divided randomly into training, selection, and testing subsets, containing 60%, 20%, and 20% of the instances, respectively.<\/p>\n Our target variable is is_fraudulent<\/b>. We can calculate the data distributions<\/a> and plot a pie chart with the percentage of instances for each class.<\/p>\n As we can see, the target variable is unbalanced, with many payments being non-fraudulent (approximately 85%), while only 15% are fraudulent. We could say that approximately 1 out of 6 payments is fraudulent.<\/p>\n The input-target correlations might indicate which factors have the most significant influence on a fraudulent transaction.<\/p>\n In this example, all variables have a positive correlation except for is_declined<\/i>. Moreover, the variable high_risk_country<\/i> has the highest correlation with the target variable.<\/p>\n<\/section>\n The next step is to set the neural network<\/a> parameters. For classification problems, it is composed of:<\/p>\n We have set the mean and standard deviation scaling method<\/a> for the scaling layer<\/a>,<\/p>\n We set\u00a0up one\u00a0perceptron layer<\/a> with 3 neurons, using<\/span>\u00a0the logistic activation function<\/a>. This layer has nine inputs, and since the target variable is binary, it has only one output.<\/p>\n The neural network for this example can be represented with the following diagram:<\/p>\n The fourth step is to set the training strategy<\/a>, defining\u00a0what the neural network will learn. A general training strategy for classification is composed of two terms:<\/p>\n The loss index<\/a> chosen for this problem is the normalized squared error<\/a> between the neural network’s outputs and the targets in the data set with L1 regularization.<\/p>\n The selected optimization algorithm<\/a> is the Quasi-Newton method<\/a>. The selected optimization algorithm<\/a> is the Quasi-Newton method<\/a>.<\/p>\n The following chart illustrates how training and selection errors evolve over the course of training epochs. The final values are\u00a0a training error\u00a0<\/strong>of\u00a00.052\u00a0<\/strong>and a\u00a0selection error\u00a0<\/strong>of 0.103, both measured in<\/strong><\/span>\u00a0NSE<\/b>.<\/p>\n The objective of model selection<\/a> is to find the network architecture with the best generalization properties, which means the one that minimizes the error on the selected instances<\/a> of the data set.<\/p>\n More specifically, we want to find a neural network with a selection error smaller than 0.103 NSE<\/b>, which is the value we have achieved.<\/p>\n Order selection<\/a> algorithms train several network architectures with a different number of neurons and select the one with the smallest selection error.<\/p>\n The incremental order<\/a> method starts with a few neurons and increases the complexity at each iteration. The following chart shows the training error (blue) and the selection error (orange) as a function of the number of neurons.<\/p>\n The selection errors achieved are similar for any number of variables; however, the smallest is 0.1007<\/b> for an optimal number of neurons of 4.<\/p>\n The graph above represents the architecture of the final neural network.<\/p>\n<\/section>\n The objective of the testing analysis<\/a> 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.<\/p>\n We can use the ROC curve<\/a>\u00a0as it is the standard testing method for binary classification projects.<\/p>\n The AUC value for this example is 0.998.<\/p>\n The following table contains the elements of the confusion matrix<\/a>.<\/p>\n This matrix contains the true positives, false positives, false negatives, and true negatives for the variable is_fraudulent<\/i>.<\/p>\n1. Application type<\/h2>\n
2. Data set<\/h2>\n
\n
Variables<\/h3>\n
Merchant Behavior<\/strong><\/h4>\n
\n
merchant_id<\/code>): Unique identifier of the merchant where the transaction occurred.<\/p>\n<\/li>\n<\/ul>\nTransaction Data<\/strong><\/h4>\n
\n
transaction_amount<\/code>): Monetary value of the individual transaction.<\/li>\navg_amount_day<\/code>): Mean value of transactions made with the card during a given day.<\/li>\nforeign_transaction<\/code>): Indicates whether the transaction was made outside the cardholder\u2019s country (Yes\/No).<\/li>\nhigh_risk_country<\/code>): Flags if the transaction originated from a country considered high-risk for fraud (Yes\/No).<\/li>\nis_declined<\/code>): Specifies if the credit card transaction was declined (Yes\/No).<\/li>\nnumber_declines_day<\/code>): Total number of declined transactions in a single day.<\/li>\n<\/ul>\nFraud Indicators & Historical Data<\/strong><\/h4>\n
\n
daily_chbk_avg_amt<\/code>): Average monetary value of chargebacks recorded per day.<\/li>\n6m_avg_chbk_amt<\/code>): Mean chargeback amount accumulated over the last six months.<\/li>\n6m_chbk_freq<\/code>): Number of chargebacks recorded in the last six months.<\/li>\n<\/ul>\nTarget variable<\/strong><\/h4>\n
\n
is_fraudulent<\/code>): Class label indicating whether the transaction was fraudulent or not.<\/li>\n<\/ul>\nInstances<\/h3>\n
Variables distributions<\/h3>\n
<\/p>\nInputs-targets correlations<\/h3>\n
<\/p>\n3. Neural network<\/h2>\n
\n
Scaling layer<\/h3>\n
Dense layers<\/h3>\n
<\/p>\n<\/section>\n4. Training strategy<\/h2>\n
\n
Loss index<\/h3>\n
<\/p>\n<\/section>\n5. Model selection<\/h2>\n
<\/p>\n<\/p>\n6. Testing analysis<\/h2>\n
ROC curve<\/h3>\n
<\/p>\nConfusion matrix<\/h3>\n
![](\"https:\/\/www.neuraldesigner.com\/images\/credit_card_positive_rates.webp\")
<\/p>\n![](\"https:\/\/www.neuraldesigner.com\/images\/credit_card_cumulative_gain.webp\")
<\/p>\n