How A National Healthcare Company Increased Forecast Accuracy by 20% in 3 Days

While they were using Machine Learning to build forecasting models, their use of the Facebook Prophet model meant that they could mostly only use time aggregation based on historical revenue information without considering other dimensional details. Relying on the Facebook Prophet model was particularly constraining since the organization needed to forecast revenue while taking into account individual product trends regional and sub-regional variables across different time dimensions.

When the company first spoke to dotData, our automated approach to building features and the power and flexibility of the platform’s handling of time-series data were of interest. The data science team wanted to achieve three critical goals: First, to boost model accuracy by leveraging more dimensions; Second, to account for the nuances of forecasting by product, region, and sub-region; and Third, to lower the impact of building, iterating, and maintaining ML models on their small data science team.

dotData’s automated feature engineering provided the ideal solution. By automatically connecting to the organizations raw data – and by automating the feature engineering process, dotData’s solution addressed several critical needs for the business. First, it allowed their data science team to connect directly to their data warehouse – bypassing the tedious and time-consuming building flat tables (feature tables) before testing ML models. An added benefit was that by connecting and manipulating raw data directly, the team could automatically explore thousands of feature hypotheses in just a few hours.

More than 67,000 features were built and evaluated. Second, the unit automatically evaluated more than 700 Machine Learning algorithms to evaluate the ideal combination of features and ML algorithms. Lastly, but most importantly, the feature engineering process explored multi-dimensional time-series data critical to building a more accurate forecasting model. The best part? One person performed all the work in under three days, resulting in a forecasting model that boosted accuracy by over 20%.

The power of dotData’s automated feature engineering has given the data science team the bandwidth to take on more projects without having to hire additional staff. The automated nature of the platform makes it possible for the team to explore data and experiment without going through lengthy, manual iterative cycles. With dotData’s powerful automation capabilities, the client is now confident that they can begin to explore more use-cases across a wide range of business needs – all with minimal impact on the valuable time of subject matter experts and without hiring an army of data scientists which is prohibitively expensive.

dotData provides an automated feature discovery platform designed to work seamlessly within the existing development process of Data Scientists, Data Engineers, and Machine Learning Engineers. Our platform connects directly to enterprise data warehouses and data lakes. It analyzes relationships among dozens of tables, hundreds of columns, and millions of rows of data to discover novel and valuable features for machine learning and advanced analytics, building the necessary data marts and feature tables automatically. ML and AI Development across the Globe rely on dotData to help them experiment with data to discover new features and accelerate the ML model building process from months to just a few days.