Incentive design for high quality disease prediction model using crowdsourced clinical data

Abstract

Abstract Clinical data mining has great potential for mining hidden patterns in the medical datasets, which can then be used to guide clinical decision making and personalized medicine. While several studies have merged medical data mining techniques with statistical analysis, their proposed mechanisms are excessively complex and are not particularly accurate for individual patients. Therefore, it is essential that a better tool is developed for disease progression and survival rate predictions. In addition, most of the medical datasets are noisy and hence any dataset needs to be cleaned before it is used for predictions. Each dataset may contain many features not all of which are useful for predictions. Therefore, useful feature selection techniques need to be employed before prediction models can be constructed. Furthermore, larger and high quality datasets typically create better prediction models. Thus, in this paper, we explore how data cleaning and feature selection techniques affect the performance of the prediction models. In addition, we develop a new incentive model with individual rationality and platform profitability features to encourage different hospitals to share high quality data so that better prediction models can be constructed. We evaluate our proposed techniques using three datasets and the results show that our proposed methods are more efficient and accurate than several existing prediction models.