A Novel Anomaly Detection Approach for Nifty Stocks using Machine Learning for Construction of Efficient Portfolio to Reduce Losses and Protect Gains

Abstract

Machine Learning is the most essential and widely utilized methodological approach these days for performing, organizing, and analyzing data using a variety of approaches to produce correct and efficient results. Machine learning methodologies are also applicable to stock markets, which have grown multi-fold in recent years, and with the amount of money involved, the possibility of manipulation is always present and has increased. Machine learning techniques can be used to detect anomalies in price behavior or price movements that are out of the ordinary. We investigated how to detect abnormalities utilizing a combination of fundamental and technical aspects in this study. We used multiple machine learning approaches to detect various types of abnormalities using fundamental and technical factors integrated into stock market data in the current proposed study work. The results have been produced utilizing a bagging strategy that included the use of a class One support vector machine, a local outlier factor, and other techniques. Companies tend to have varying valuations across sectors, but generally follow a range for a specific industry type; hence the data sets have been divided by sector. The results were segmented into the industry type, such as banking, cement, and energy. The portfolio has been built using anomaly scores. The method assisted in the removal of equities from the portfolio, avoiding losses and preventing profits from eroding. The mean absolute error has been determined to be 10.22%, which enhances the whole system's ability to detect anomalies.