A Data-Driven Approach to Understanding the Impact of Covid-19 On Dietary Habits Amongst Bangladeshi Students

Abstract

The advent of COVID-19 has brought upon behavioral changes in food habits and overall nutritional lifestyles. It is necessary for nutritionists to analyze how the crisis affected human behavior, particularly general well-being. The purpose of this research is to see how quarantine affected food habits amongst Bangladeshi youngsters. A cross-sectional survey consisting of a three-part questionnaire (personal details, pre-COVID habits, and during COVID habits) was done on 230 students. Exploratory data analysis was done to summarize the resulting dataset. Afterward, K-means clustering was used to find the natural groupings that exist in this dataset. Later on, elbow and silhouette methods were applied to determine the number of optimal clusters so that machine learning classifiers could be run on them. Finally, SHAP was implemented to identify the features' importance. After clustering the data, the silhouette method identified 5 as the optimal cluster number to try the models out on. Out of all the models utilized, Extra Tree Classifier worked the best since it scored the highest in all the evaluation criteria. SHAP revealed that the habit of not eating carrots had the most significant impact in predicting the membership of a cluster class. These findings demonstrated that although there was indeed a negative edge to the crisis, there were slits of positive outcomes among nutritional behaviors which could be leveraged for a healthier societal change. Some of the values are within the consensus (reducing intake of fruit juice) while others are novel (smoking patterns being unchanged). These ideas will help in creating a broader idea as to how nutrition can take a nosedive and can be managed using data.