Predict the prevalence and incidence of Parkinson’s disease using fractal interpolation-LSTM model

Abstract

The investigation of the prediction of disease population is a noticeable exploration topic in the field of sciences. As a type of neurological disease, the incidence and prevalence of Parkinson’s disease are still difficult to accurately study. In this paper, a method is proposed to forecast the number of incident cases (NumIn), incidence rate (InRa), the number of prevalent cases (NumPr), and prevalence rate (PrRa) of Parkinson’s disease in ten countries selected. Using past data on the incidence rate, the number of prevalent cases, and the prevalence rate from 1990 to 2019, three types of fractal interpolations with different fractal dimensions are constructed for reconstructing the past data, where the vertical scaling factors are determined by the method proposed in this article. Then, the Long Short-Term Memory (LSTM) model is employed to forecast the values of NumIn, InRa, NumPr, and PrRa with Parkinson’s disease in 2020. Meanwhile, the autoregressive integrated moving average model is used to predict the values compared with the LSTM model. The evaluation metrics employed for error calculation include the root mean square error and the coefficient of determination (R2). According to the proposed optimal criteria, the best predicted results are the average of three types of prediction values based on the LSTM model by analyzing and comparing eight predicted results.