Data Leakage Behavior Recognition Based on Cyclical Learning Rate based Long-Short Term Memory

The data leakage behavior are malicious activities which considered through an authorized member inside an organization. The insider threats are a challenging one for public and private organizations which cause numerous damage to organizations. The existing techniques in the insider threat focuses on attack detection. This paper proposed a Cyclical Learning Rate based Long-Short Term Memory (CLR based LSTM) for data leakage behavior recognition. The CERT dataset is employed which comprises 32,770,220 records of malicious and normal users. The dataset is preprocessed through data cleaning, transformation and reduction. The preprocessed data features are extracted then the CLR based LSTM is applied for classification. The accuracy, precision, f1score and recall metrics are utilized to measure the model performance. The proposed model achieved 98.6% of accuracy, 98.2% of precision, 97.7% of f1-score, 97.3% of recall when compared to existing techniques like LSTM based Autoencoder, Variational Autoencoder (VAE), Multi-Fuzzy Classifier and Adaptive Deep Neural Network (AD-DNN).