Automatic identification and explanation of root causes on COVID-19 index anomalies

This paper reports a method for automatically identifying, analyzing and explaining anomalies in different indexes of COVID-19 crisis using Artificial Intelligence (AI) based techniques. With systematic application of News sensor, language detection & translation, Keyword-based extraction of COVID-19 indexes, Convolutional Neural Network (CNN) based anomaly detection, and Natural Language Processing (NLP) based explanation methods, this paper demonstrates a methodological solution for strategic decision makers to make evidence-based policy decisions on COVID-19 (in multiple dimensions like Travel, Vaccine, Medical etc.). Firstly, COVID-19 related News is fetched from multiple sources in multiple languages. Then, AI-based language detection and translation process automatically translates these News and posts in real-time. Next, COVID-19 related News and posts are segregated in multiple groups using pre-defined keywords for creation of multiple indexes. Lastly, CNN based anomaly detection identifies all the anomalies on multiple COVID-19 indexes with NLP-based explanations. A standalone decision support system was developed that implemented the presented method. This decision support system allows a strategic decision-maker to comprehend “when, where, and why there are fluctuations in COVID-19 related sentiments on a particular dimension”. Method was validated with Tweets from 15/072021 to 24/05/2022 resulting in automated generation of 5 COVID-19 indexes and 69 anomalies with explanations. In summary, this method of anomaly detection on COVID-19 indexes presents:•An explicit, transferable and reproducible procedure for detecting anomalies on multiple indexes of COVID-19 in multiple languages•It helps a strategic decision maker to comprehend the root-causes of anomalies in COVID-19 related travel, vaccine, medical indexes•The solution developed using the presented method allows evidence-based strategic decision-making COVID-19 crisis using AI, Deep Learning and NLP [Display omitted]