Predictive analysis can be attributed to the development of popular models of modeling, machine learning, and data mining to address complex real-world problems and is intended to enable computers to help humans analyze complex and large data sets. predictive analytics approaches have largely emerged for application in complex disease areas that have involved significant research such as, diseases or patients classification, clustering, prediction, as well as drugs discovery and repositioning. neurodegenerative diseases are a group of complex diseases caused by chronic and progressive neural degeneration. early detection and progression monitoring with objective measures of these diseases can play an important role in mitigating a hard situation and in aiding treatments.indeed, after an in-depth analysis of the different modalities resulting from risk, biomedical and neuroimaging measures, we identify and present the speech and handwriting which are present in parkinson's disease (pd), and represent an important non-invasive asset for early appearance. thus, we proposed frameworks using shallow learning to early detection of pd. the first uses double parallel feed-forward neural network to learn pattern of speech. this network is optimized using evolutionary algorithm in number of hidden nodes and signal features. in the second, a new coding architecture x is introduced to improve the evolutionary algorithm for obtaining an optimal parallel network to learn pattern of speech and handwriting. the experiments were conducted on several models to measure the accuracy and efficiency of the proposed approach, and the results were promising, outperform state-of-the-art speech and handwriting algorithms.