Anomaly Detection in Time Series Data of Sensex and Nifty50 With Keras

Citations Over Time

Abstract

Anomaly detection problem for time series refers to finding outlier data points relative to some standard or usual signal. A price action that contradicts the expected movement of the stock market is called an anomaly. Few anomalies appear only once and disappear, but there are some that appear consistently throughout historical chart analysis. Traders and investors are benefitted from these unusual market behaviors to find opportunities throughout the stock market. While there are plenty of anomaly types, we have focused only on the most important ones such as unexpected spikes i.e., sudden rise in the prices, drops i.e., sudden fall in the prices, trend changes and level shifts. Using Long Short-Term Memory (LSTM) we have predicted the points where there might be a possible anomaly in the time series. Using the time series of SENSEX and NIFTY50 over the last 25 years we have predicted the possible anomaly in the time series. We have first imported the libraries, loaded and inspected the data of both SENSEX and NIFTY50. We have then processed the data and then created and split the data into test and train dataset. Then we have built an LSTM Autoencoder. An LSTM Autoencoder is an implementation of an autoencoder for sequence data using an Encoder-Decoder LSTM architecture. Once fit, the encoder part of the model can be used to encode or compress sequence data that in turn may be used in data visualizations or as a feature vector input to a supervised learning model. We have then evaluated the model and then detected the possible anomalies in the entire dataset.

Related Papers

• → Improved autoencoder for unsupervised anomaly detection(2021)119 cited
• → Why is this an anomaly? Explaining anomalies using sequential explanations(2021)26 cited
• → Toward Unsupervised 3d Point Cloud Anomaly Detection Using Variational Autoencoder(2021)22 cited
• → Identification of outliers in pollution concentration levels using anomaly detection(2016)6 cited
• → Toward Unsupervised 3D Point Cloud Anomaly Detection using Variational Autoencoder(2023)1 cited