smjajoo / pima-data-modeling-and-analysis

This project looks at the effectiveness of SICE imputation technique in supporting binary classification performance of the Logisitic Regression in the context of decision support for the healthcare sector where accurate predictive models have the potential to improve patient outcomes by promoting access to care. Processing methodologies are assesed with reference to the relevant Accuracy and False Positive metrics. Analysis and comparison of the results suggest that imputation alone can likely impact classifier performance only marginally in situations where the data set is of sufficient size.

PIMA-data-modeling-and-analysis

This project looks at the effectiveness of SICE imputation technique in supporting binary classification performance of the Logisitic Regression in the context of decision support for the healthcare sector where accurate predictive models have the potential to improve patient outcomes by promoting access to care. Processing methodologies are assesed with reference to the relevant Accuracy and False Positive metrics. Analysis and comparison of the results suggest that imputation alone can likely impact classifier performance only marginally in situations where the data set is of sufficient size.

1. Project's Title:

   Modelling, Analysis & Classification of the Pima Indians Diabetes Mellitus Dataset

2. Project Description:

   • Intro of project: This project looks at the effectiveness of different imputation techniques in supporting binary classification performance of the Logisitic Regression, J48 decision tree and KNN classifiers in the context of decision support for the healthcare sector where accurate predictive models have the potential to improve patient outcomes by promoting access to care. Processing methodologies are assesed with reference to the relevant Accuracy and False Positive metrics. Analysis and comparison of the results suggests that imputation alone can likely impact classifier performance only marginally in situations where the data set is of sufficient size.

   • Technologies used and reason: RStudio RStudio allows users to develop and edit programs in R by supporting a large number of statistical packages, higher quality graphics, and the ability to manage your workspace.

   • Faced challenges and features to hope to implement in the future: Overall the pre-processing approaches utilised were demonstrated to have minimal effectiveness in promoting classifier performance. A weakness of this study was to limit exploration of pre-processing approaches largely to imputation techniques. More impactful analysis might have been obtained if the employed approaches had gone further by imposing more rigorous feature selection or transformation techniques. An obvious point of further investigation would be in assessing the value of imputation at all, it may simply be situationally appropriate when the size of the data is constrained to the point where removing data will reduce the statistical power of the data set but further work would have to be undertaken to demonstrate this.