Development and Performance Analysis of Machine Learning Methods for Predicting the Occurrence of Constipation and its Risk Factors Among College-aged Girls

The present study sought to determine which model was most useful for predicting functional constipation (FC) in college-aged students by examining the applicability of multiple models and evaluating the forecasting accuracy of prediction methods, including regression-based models and machine learning models. This observational descriptive study involved 300 college girls from Kolkata, West Bengal, India, who were randomly chosen using social media (Linkedin,WhatsApp and Face book) and ranged in age from 18 to 25 years. The survey was carried out using an online, standard questionnaire that had been pre-tested. The obtained data were entered into a Microsoft Excel Worksheet (Redwoods, Washington, USA: Microsoft) and reviewed for elimination errors.19 attributes were selected for prediction study. Weka version 3.8.0 software was used for predictive modeling, performance analysis, and the building of FC prediction system. The predictive models were then developed and contrasted using 5 different models as a classifier. We divided our data into training and test datasets, which comprised 70% and 30% of the total sample, respectively, at random for each investigation. Out of 300 occurrences, 96.00 % were correctly classified, while only 4 % were wrongly classified, with a Kappa value of 0.875, and a root mean squared error of 0.19. The model's accuracy was 96.3% weighted precision, 96% true positives, 0.05% false positives, 0.961 F measure, and 0.994ROC(receiver operating characteristic curve).Here 6 different evaluators were used and surprisingly they all predict Bristol's Stool consistency Scale as the number 1 predictor of FC among college girls. Again ‘Pain and discomfort in abdomen’ remains second predictor according to all selected evaluators. Thus, it can be confirmed that ‘Bristol's Stool consistency Scale’ and the ‘Pain and discomfort in abdomen’ are the two significant predictor of FC among college going girls. This machine learning model-based automated approach for predicting functional constipation will assist medical professionals in identifying younger generations who are more likely to experience constipation. Additionally, predictions can be made quickly and efficiently using sociodemographic and morbidity parameters. For further follow-up and care, at-risk patients can be referred to consultant physicians. This will lessen the burden of gastrointestinal-related morbidity and mortality among the younger population.