Feature Selection: Role in Designing Smart Healthcare Models

With the leveraging fields of data mining and artificial intelligence, the data is growing day by day in an exponential manner. In health care sector, huge amount of biomedical data are generated from online hospital management applications, online sites, biomedical devices and sensors, and from various other electronic devices in an innumerous manner. There is a huge demand in the health sector to store this data so that they can be analyzed for future predictions. The process involves storing of high dimensional data and also very high volume of data (“Big Data”), which later needs to be processed to extract the desired information. The important attributes of the data needs to be identified, and then a subset is to be generated which can help in training the prediction model classifiers. This big data needs to be preprocessed to eradicate the unrequited attributes, identifying the essential attributes, then filtering out the noise (irrelevant attributes) from it and minimize its size without affecting its quality, i.e. after filtrating the attributes, there should not be any required attribute missing which will affect the performance of the predictive models. This task of identifying patterns, identifying relevant attributes for prediction of diseases and selecting min. no. attributes from the huge data set so as to achieve the results from predictive models with minimum time and cost is very challenging and requires a lot of expertise. This book chapter explains the importance of feature selection and feature creation related to biomedical data. Actually first a set of features have to be created which will study the hidden behavior and patterns and these features will then in turn be used to identify patterns and assist in prediction of diseases. Feature identification is again a rigorous task and involves lot of expertise. This chapter gives an insight into why feature selection is essential in designing the smart healthcare predictive models for real time data.