Improving Medical Diagnosis Using Missing Data Treatment Techniques: A Case Study on Thyroid Data

In this paper, we focus on some advanced input techniques that deal with missing data from thyroid medical datasets such as maximum expectation (EM), multiple sequential input (MICE), etc. It is demonstrated that EM and MICE had a significant positive impact on prediction accuracy compared to their traditional counterparts (i.e. direct and ensemble inputs). Moreover, EM significantly enhanced predictions derived from random forests in a manner consistent with previous findings of additive predictive power for clinical variables only, confirming the promise of EM (and MICE) to provide a significant fundamental improvement in modeling in complex medical models, in the case of a comprehensive analysis of phenotypes similar to heart failure. In conclusion, the results of the present study indicate that advanced input methods increase the accuracy of diagnosis and therapeutic predictions in patients with thyroid diseases.