Comparative Analysis of Machine Learning Classifiers Using a Non-Stationary Dataset
Raghav Uparkar, Sudhanshu Maurya, Prapti S Barde, Shriya Kalbande, Geeta Naidu, Sagarkumar S. Badhiye · 2024
Non-stationary data sets related to time series have been a major concern of the researchers due to their venerable characteristics. This research study involves a data analytical approach using different techniques to check whether the time series is stationary or not. A real-time data set from the Yahoo finance site of the National Stock Exchange of India (NSEI) is used. Initially, the Visual test, Summary Statistics, ADF test, and KPSS test were used to test the hypothesis of non-stationarity. After passing all of these tests, an innovative approach is employed to calculate a binary goal value based on the daily buy or sale of the stock. The logistic regression approach first investigates two models without and with random effects to determine the consistency of the binary target value. Machine learning classifiers are used to compare various metrics such as accuracy, recall, precision, F1 score, and execution time. This research study aims to develop a more effective data analytical approach for non-stationary data sets.