Machine Learning: Bridging Statistics and IT
What do you do with millions of data points from multiple data sources that need to be categorized, coded, and analyzed? And in real time? And within a tight budget? Coding and categorizing it all could take years to complete. To say nothing of the cost. Machine learning provides the remedy.
With machine learning, we manually review and categorize subsets of available data. We then train the system using those subsets thru the latest machine learning techniques to automatically code and categorize raw data. We can recalibrate the process over time to deal with difficult data patterns and changing requirements.
Machine learning lets us set up an infrastructure that we can receive and review massive amounts of data, and quickly spot, analyze, and report on trends.
Westat uses advanced methods for solutions
Westat has harnessed the power of statistics and IT to solve data management challenges. We’ve developed a multipronged approach using natural language processing, machine learning methods, and statistical algorithms. Our toolkit draws on neural network and support vector machine methods, latent semantic indexing, and other advanced statistical methods.
Good prognosis for processing hospital survey data
Machine learning is a great tool to use when processing large-scale, longitudinal data. Take, for example, a survey that provides national data on inpatient hospital care. Westat collects millions of medical claims records each year for the survey. The data is sent to us via a secure site.
Using machine learning, we developed a system to automatically categorize payer type based on the payer name listed in the records:
- We built dictionaries to preprocess the raw data into usable inputs.
- We trained the system with that preprocessed data and used the resulting “models” to code new data.
- We set up an infrastructure for data management to review, check quality, annotate, and update results.
Our system has processed tens of millions of records, something that previously required intensive manual labor. We also developed a system to streamline data quality control so that manual review is reduced by 80%. This allows data management staff to focus on resolving more difficult data issues.