Feature Selection Techniques to Enhance Prediction of Clinical Appointment No-Shows Using Neural Network

The issue of no-shows is a significant concern, as it results in many patients missing their appointments at outpatient clinics worldwide, often without any prior cancelation. This leads to inefficiencies in terms of idle resources and wasted capacity. Hence prediction models are needed to anticipate whether a patient will attend their scheduled appointment. The effectiveness of a predictive model in predicting clinical appointment no-shows is heavily influenced by the features used in the model. To address this issue, the present study compares various feature selection techniques in order to enhance the accuracy of prediction. Univariate Selection, Recursive Feature Elimination, Random Forest Classifier, and Reciprocal Ranking are the feature selection techniques utilized in the current study. These techniques are applied prior to building a neural network model using a Multilayer Perceptron to predict clinical appointment no-shows. The study employed the scikit-learn library in python for model implementation, and the performance of each model is evaluated using performance measures, including Accuracy, Specificity, Sensitivity, Precision, F-measure, Matthews Correlation, Log Loss, and Area under the Curve. Feature selection methods demonstrated excellent performance by reducing the number of variables while maintaining the predictive accuracy across all models. Consistently, the most critical features across all models were the Patient Trust, Appointment Type, Reminder Message, Lead Time, and Missed Appointment History.