Utilizing valuable historical information and using techniques like recommendation, forecasting via data modeling to provide domain specific decisions. This reduces decision-making time and through advance deep learning techniques we can utilize the existing domain experience to improve day-to-day decision-making. We have built adaptive random forests, similarities models, Bayesian modeling of existing data based on R, Python, Weka and using big data sets on Hadoop/ Spark etc.
Case Study: To create a recommendation system for helping doctors to prescribe medicines/drugs to the patients.
Huge amount of unstructured patient consultation data in the form of doctors’ notes, patient symptoms, diagnosis, psychical attributes of the patient like age, weight, height, BMI and basic medical condition check like diabetes, Blood pressure, thyroid. Along with the case details the doctors’ treatment methodologies and the medicines prescribed.
A simple approach can be similarity based model like KNN. But after doing good research on the data, we went with random forest model which captures the essence of the data. After the feature engineering is performed, we modeled against the medicines to be recommended. After many rounds of fine tuning, we were able to get great results. The models were converted into PMML form, which aids faster deployment and real time advantage.