Model-based clustering in longitudinal data: a data-mining approach

Consider an experiment in which p measurements are observed in each of n treatments for an observation . This is a longitudinaldata experiment . In this regard , we have some matrices as observations . The clustering algorithm as a data-mining approach canhelp us to find some relationship between data and categorize them in some classes. In classical clustering methods such asK-means or hierarchical clustering , we try to categorize the observations into one of several groups based on some vector-valuespredictors . In this paper , we assume the predictors are as matrices and develop a clustering method based on finite mixtures ofmatrix-variate normal distribution . In a real experiment on Parkinson's disease patients , the performance of the model in theclustering of data is studied .

کلیدواژه ها:

Matrix-variate normal ، Model-based clustering ، data-mining ، Misclassification error rate ، Parkinson's disease