Predictive Analytics and Modeling Employing Machine Learning Technology: The Next Step in Data Sharing, Analysis, and Individualized Counseling Explored With a Large, Prospective Prenatal Hydronephrosis Database
Journal Articles
Overview
Research
Identity
Additional Document Info
View All
Overview
abstract
OBJECTIVE: To explore the potential value of utilizing a commercially available cloud-based machine learning platform to predict surgical intervention in infants with prenatal hydronephrosis (HN). MATERIALS AND METHODS: A prospective prenatal HN database was uploaded into Microsoft Azure Machine Learning Studio. Probabilistic principal component analysis was employed for data imputation. Multiple clinical variables were included in two-class decision jungle and neural network for model training, using surgical intervention as the primary outcome. Models were scored and evaluated after a 70/30 split of the data. RESULTS: A total of 557 entries were included. The optimized model (decision jungle) achieved an area under the curve of 0.9, accuracy of 0.87, and precision of 0.80, employing a threshold of 0.5 to predict surgery. Average time to train, score and evaluate the model was 5 seconds. The predictive model was deployed as a web service in 35 seconds, generating a unique API key for app and webpage development. Individualized prediction based on the included variables was deployed as a web-based and batch execution Excel file in less than one minute. CONCLUSION: This cloud-based ML technology allows easy building, deployment, and sharing of predictive analytics solutions. Using prenatal HN as an example, we propose an opportunity to address contemporary challenges with data analysis, reporting a creative solution that moves beyond the current standard.
