To name seven (7):
(1) User segmentation using a proprietary variation of a hierarchical DBSCAN clustering algorithm of high-dimensional data with novel distance [quasi] metric, based on hubness analysis;
(2) Leveraging the above in email targeting and invoking multi-armed bandit testing methodologies for email timing, frequency, and content, using decreasing-epsilon strategy;
(3) Modeling predicted underwriting criteria with a binary approval odds classification algorithm;
(4) Using a dynamic panel data, fixed effects model to predict the effect of changes in credit reports on user credit score;
(5) Employing an Autoregressive Integrated Moving Average (ARIMA) with optimized Akaike Information Criterion exploits to predict future revenue and growth (lagged results led to average error bounds of only 5 percent; cross-validation results were even stronger, though I was conservative in guaranteeing 7 percent error, on average);
(6) Refining a multiverse (context-aware) recommendation engine as an n-dimensional tensor (rather than the typical two-dimensional user-item matrix) for partner product recommendations, using High-Order Singular Value Decomposition to solve;
(7) Invoking a Convolutional Neural Network framework with a novel architecture and results of a Fourier Transform as input to classify dental x-rays and highlight to the dentist which teeth require fillings (after approximately two months, the model reached ~95 percent accuracy - in terms of actual agreement by dentists using the app - with F1 score in cross-validation performing on par). Review collected by and hosted on G2.com.