0 ×

Keras_​Autoencoder_​for_​Fraud_​Detection_​Training

Workflow

Keras Autoencoder Architecture Data Preprocessing Training the Autoencoder Train Keras Autoencoder for Fraud DetectionCreate a training set, validation set for threshold optimization and networkvalidation setBuild Keras AutoencoderCalculate reconstruction errors (MSE) and optimize the thresholdEvaluate the model performace using the validation set for thresholdoptimization and the optimal threshold Read credit card dataTop:Class = 02/3 of negativesfor training1/3 of negativesand all positivesfor validationMin-max normalizationOutput 14Activation: tanhwith activity regularizer L1=10e-5Output 7Activation: reLUwith activity regularizer L1=10e-5Define outliers bythe best threshold forreconstruction errorOutlier => 1No outlier => 0Train with Loss function=MSEOptimizer=AdamOutput 7Activation: tanhwith activity regularizer L1=10e-5Output 30Activation: reLUwith activity regularizer L1=10e-510 % for validationWrite creditcard_autoencoder.h5Normalizerfor deployment File Reader Row Splitter Partitioning Concatenate Normalizer Keras Input Layer Keras Dense Layer Keras Dense Layer Rule Engine Number To String Keras NetworkLearner DL Network Executor Normalizer (Apply) Keras Dense Layer Keras Dense Layer Partitioning Keras NetworkWriter Scorer (JavaScript) ThresholdOptimization Model Writer Normalizer (Apply) Keras Autoencoder Architecture Data Preprocessing Training the Autoencoder Train Keras Autoencoder for Fraud DetectionCreate a training set, validation set for threshold optimization and networkvalidation setBuild Keras AutoencoderCalculate reconstruction errors (MSE) and optimize the thresholdEvaluate the model performace using the validation set for thresholdoptimization and the optimal threshold Read credit card dataTop:Class = 02/3 of negativesfor training1/3 of negativesand all positivesfor validationMin-max normalizationOutput 14Activation: tanhwith activity regularizer L1=10e-5Output 7Activation: reLUwith activity regularizer L1=10e-5Define outliers bythe best threshold forreconstruction errorOutlier => 1No outlier => 0Train with Loss function=MSEOptimizer=AdamOutput 7Activation: tanhwith activity regularizer L1=10e-5Output 30Activation: reLUwith activity regularizer L1=10e-510 % for validationWrite creditcard_autoencoder.h5Normalizerfor deployment File Reader Row Splitter Partitioning Concatenate Normalizer Keras Input Layer Keras Dense Layer Keras Dense Layer Rule Engine Number To String Keras NetworkLearner DL Network Executor Normalizer (Apply) Keras Dense Layer Keras Dense Layer Partitioning Keras NetworkWriter Scorer (JavaScript) ThresholdOptimization Model Writer Normalizer (Apply)

Download

Get this workflow from the following link: Download

Nodes

Keras_​Autoencoder_​for_​Fraud_​Detection_​Training consists of the following 34 nodes(s):

Plugins

Keras_​Autoencoder_​for_​Fraud_​Detection_​Training contains nodes provided by the following 7 plugin(s):