Mining Good Sliding Window for Positive Pathogens Prediction in Pathogenic Spectrum Analysis

Positive pathogens prediction is the basis of pathogenic spectrum analysis, which is a meaningful work in public health. Gene Expression Programming (GEP) can develop the model without predetermined assumptions, so applying GEP to positive pathogens prediction is desirable. However, traditional time-adjacent sliding window may not be suitable for GEP evolving accurate prediction model. The main contributions of this work include: (1) applying GEP-based prediction method to diarrhea syndrome related pathogens prediction, (2) analyzing the disadvantages of traditional time-adjacent sliding window in GEP prediction, (3) proposing a heuristic method to mine good sliding window for generating training set that is used for GEP evolution, (4) proving the problem of training set selection is NP-hard, (5) giving an experimental study on both real-world and simulated data to demonstrate the effectiveness of the proposed method, and discussing some future studies.