Loading...

Improving Outpatient Appointment System Using Machine Learning Algorithms and Simulation

Sadeghi, Niloufar | 2020

372 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 53069 (01)
  4. University: Sharif University of Technology
  5. Department: Industrial Engineering
  6. Advisor(s): Khedmati, Majid; Najafi, Mehdi
  7. Abstract:
  8. The outpatient clinics have some features that make them different from other types of the clinic. Something that must be paid attention is adherence to appointments by outpatients, which have an undeniable effect on the productivity of the Outpatient Appointment System. In most of the outpatient clinics, outpatients are supposed to schedule their appointment in advance. As a result, if patients do not come to the health center based on their scheduled appointment, and they do not inform the health center about their absence, not only will the clinic become deprived of the opportunity to utilize its resources efficiently, but also healthcare access for other patients will reduce. The rate of no-show has a significant impact on the scheduling process, and if the rate is high, there may be many disruptions to the scheduling programs. The new approach to scheduling patients based on no-show rates is to consider this rate individually based on each patient's characteristics. This assumption is very close to the real world because factors such as age, gender, level of education, distance to the center, the interval between scheduled booking dates and scheduled appointments, and previous no-show history are considered in the prediction model.The first novelty of the research is that extracting knowledge in this research is coming from applying machine learning algorithms on outpatients of an eye hospital in Tehran; therefore, we are working on a real case study. Another contribution of this study is the implementation of some methods to come up against the imbalanced data-set of the research. In this problem, the proportion of the majority class to the minority class is high, and most of the conventional methods are designed for balanced data. As a result, in imbalanced data, the prediction will be biased toward the majority class, and minority class is at risk of misclassification. There are different methods in the literature, such as preprocessing techniques, cost-sensitive learning, and ensemble methods for dealing with imbalanced data. In this study, a combination of preprocessing techniques such as SMOTE, Borderline SMOTE, ADASYN, SMOTEENN and machine learning algorithms such as logistics regression, random forest, adabost and gradient boosting have been used. Based on the output of the models, random forest algorithms and gradient grading showed the best performance and logistic regression algorithm showed the weakest performance. Four methods, SMOTE, Borderline SMOTE, ADASYN and SMOTE Tomek, have performed almost the same function. However, the Borderline SMOTE method performs slightly better than other models in the Accuracy criterion, while performing the same function on other criteria as the other methods. Finally, the results of the predictive phase in the simulation phase have been used to optimize the policies for determining the turnover in excess of the capacity and determining the sequence of patients. Since the probability of each patient's absence is predicted separately, the patients' appointments are arranged in such a way that the probability of the presence of both patients at the same time and the absence of both patients in the same slot is minimized
  9. Keywords:
  10. Discrete Event Simulation ; Machine Learning ; Outpatient ; No-Show Rate ; Outpatient Appointment System ; Imbalanced Data

 Digital Object List

 Bookmark

No TOC