Sliding Window-Based Anomaly Detection

Anomaly detection is a significant research problem requiring the ability to recognize patterns in data that differ from their original pattern. There are many machine learning and deep learning models for anomaly detection. We are working with k-Nearest Neighbours(kNN), Support Vector Machine (SVM), Decision Tree (DT), Long Short Term Memory (LSTM), 1D Convolutional Neural Network with LSTM (CLSTM), and Encoder-Decoder LSTM (ED LSTM). In this work, we have used sliding windows employed with these models. This technique puts the data into slices of the same size and processes them sequentially. We have applied different window sizes [24, 48, 96, 196]. We have compared the performance of our models with or without sliding window-based techniques, analyzed the results, and found that deep learning models’ performance increases after applying the sliding window technique. The ED LSTM model has shown the highest performance increase with a sliding window for the Okhla Phase2 dataset; the performance gain is approx. 10 percent.