Analysis of anomaly and novelty detection in time series data using machine learning techniques

The process of identifying new patterns or peculiarities that exist in the regular time series data is called time series novelty detection or anomaly detection. Although it is one of the most difficult data mining areas, it is gaining popularity due to its quick application to real-world problems. This research proposes a novel way to detect time series novelty using ML algorithms. An usage of suggested ML techniques to find outliers in time series data has increased recently. Using a dataset from Stack Overflow, this research investigates the use of machine learning for anomaly and novelty identification from time series data.The initial data preparations were dealing with missing values, examining tags of datasets and data visualisation through individuals’ words and words using the following assessment metrics: MAE=0. 0629, RMSE=0. 089, and MSE=0. 007. The performance of the second-best model, ARIMA, yielded an MAE = 0. 068, RMSE = 0. 0936 and MSE = 0. 008. The lowest accuracy for this task was witnessed with the Decision Tree Regressor since its error rates were the highest. The results confirm the suitability of the Random Forest Regressor in increasing accuracy in time series data with a special focus on novel and abnormal data point detection while emphasising the significance of the model choice.