A New-Style Random Forest Diagnosis Model for Alzheimer's Disease

In this paper, a new algorithm named cost sensitive and random forest -weighted algorithm (αCSRF) is proposed as a diagnostic model for Alzheimer's disease (AD). In the new algorithm, cost-sensitive learning is introduced into random forest algorithm, and weighted sum terms of information gain ratio and misclassification cost decline ratio are constructed. In this model, [18F] AV1451 Tau-PET imaging data of areas of interest in the brain were selected as biological markers to classify disease course into three categories: normal control (NC), mild cognitive impairment (MCI) and AD. Experiment proved that this model is a dynamic model that can calculate the misclassification cost and the classification accuracy by adjusting the parameters. According to the actual requirements, the weighting parameters can be selected to obtain a model with better comprehensive performance. In this experiment, when the parameter is 0.6, the total error cost of the misclassification is quantified to 46.9, and the accuracy is 81.6%, which is the optimal comprehensive performance. Compared with other methods, the algorithm (αCSRF) proposed in this paper is more flexible, more practical and more robust.