At DFX-5, we identify specific outcomes (at the most granular level or basic unit of measurement) through data mining. As opposed to forecasting, in predictive modeling, there is little to no direct data available, no recognizable pattern or unique outliers, and no historic data that is specific to the desired outcome. Therefore, our solutions must mine data from their own created data points.

 Open Source Predictions

 Open source is the future of a shared information world.  Ubiquitous computing, in which all elements in our environment are connected to the internet and share all data is rapidly becoming a reality. The power of upcoming 5G networks are set to accelerate the interoperability of devices with each other. Imagine you are driving down main street and your vehicle is in contact with all other vehicles in its vicinity, the traffic lights are aware of your positioning, and based on your route, pedestrian crosswalks automatically create safe paths for crossing. Imagine it starts to rain and a sensor is triggered to determine road-surface conditions. If the conditions worsen, then the sensor automatically diminishes dynamic speed limits or modifies traffic flows. The idea is that devices will connect with each other in open source. At DFX5 we use these open sources to gain visibility about market conditions. Based on business rules as defined by you, we then create rules and conditions to automatically act based on market conditions. The results are validated and then sent back to the algorithm for analysis. The algorithm will use the real-time conditions and results of previous conditions to make accurate predictions of what will happen next. The results are then validated and fed back into the algorithm, ad infinitum. Thus, creating an ever improving prediction engine that started without any specific data point and just relied on open source information.

Industry Specific Predictions

There are situations, particularly in highly technical enterprise environments, in where no open source is available for consumption and no other private or proprietary data source is available. In these scenarios a predictive engine must literally start from scratch. While not ideal, even basic enterprise data is better than no data; yet a predictive model can be built without historic data points. At DFX5 we train our predictive models by using a multi-disciplinary approach that generates a wide set of algorithms that predict a wide-ranging set of parameters. The more algorithms and the more parameters the better. While at first, we may suffer from bias or variance, the purpose is to find correlations between the metrics and the results. Algorithms have become much more efficient at finding these correlations than humans. Thus, we use machine learning tools like boosting, bagging, and regularizing the results. The idea is to find the most efficient correlation out of the realm of identified correlations. Once we have identified the parameters that best assist in the prediction of an outcome, we focus on tuning or optimizing those algorithms. When tuning we focus more on the results of the prediction and seek to diminish the ‘loss function’ or the difference between our prediction and the actual result. Until the point in which the algorithm can predict with accuracy results without the need to ingest historic data points.

Predictions are generated using very-complex algorithms like K-nearest neighbor, Naïve-Bayes, Mean Variance Standardization, Neural Networks, and many more to accurately predict the result. Typical predictions answer these types of questions:

• Could a random stranger, located in a random city, buy my product brand?
• How would this medication react in a unique clinical situation?
• Is the credit worthiness of this client suitable for a specific type of loan?
• Which potential investment is best suited for our portfolio?