Iodine Software hired Aviana to improve the accuracy of predictive models used to correlate data from patient medical records with the recorded clinical diagnosis. The models use both structured and unstructured data as predictors.
Aviana used machine learning techniques of feature selection, model segmentation and logistic regression to improve the performance of one of Iodine’s most challenging predictive models by 400%.

Subsequent to the first engagement, Aviana worked for two follow-on projects with Iodine Software:

  • Development of models to measure the correlation between the estimated length of a hospital stay and the total number of procedures, and the relationship between the length of stay and the occurrence of a particular medical condition. This addressed whether some of the enhanced model predictors capture or are a surrogate for length of stay, and therefore would not be effective for early stage scoring for occurrence of a specific medical condition.
  • Predictive models for additional specific medical conditions, using the techniques developed in the first project.

In addition to the above engagements, Aviana provided hands-on mentoring of best-practices in predictive modeling to Iodine Software’s technology staff.

Aviana also evaluated data science tools where we compared Iodine’s chosen modeling tool, H2O, to IBM’s SPSS Modeler and the open source data mining tool, KNIME.