In Part 1 of 2 of this segment, we saw the limitation of using a traditional prediction model like logistic regression to correctly classify two colors in a noisy dataset. Then, we built our own neural network structure, initialized parameters, and computed the forward propagation activation functions.

In this Part 2 of 2, we will complete the build of our neural network model to better classify the color dots in our original dataset. Specifically, we’ll review the cost function, backward propagation, parameters updates, and the final model assembly/prediction.

We discussed earlier the importance of minimizing cost in our neural network…

This two-part article takes a more holistic, overarching (and yes, less math-y) approach to building a neural network from scratch. Python for completing the network is also included in each of the 7 steps.

**Part One: **(1) Define the network structure, (2) Initialize parameters, and (3) Implement forward propagation. **Part Two****: **(4) Estimate cost, (5) Implement backward propagation, (6) Update parameters, and (7) Make predictions.

Data is often non-linearly distributed, or contains unusual boundaries for which traditional classification models can’t differentiate very well. …

Patient length-of-stay (LOS) defines the time a patient spends admitted at a healthcare facility. While the implications of LOS are vast, here we will focus on its positive correlation with the following negative and interrelated outcomes in a hospital setting:

- Greater and unexpected hospital resource demands;
- Potential for patients to acquire healthcare-associated infections (HAIs), which;
- Increase the payor’s likelihood to deny claims, putting hospitals at financial risk

Therefore, the ability to predict LOS could be advantageous by allowing hospitals time to better allocate resources to meet that unexpected demand, thereby improving health outcomes for patients, lowering costs, and also satisfying…

Current Procedure Terminology codes, or CPT codes, are an especially nuanced and challenging variable type to pre-process. For starters, they are alphanumerical, non-ordered, combinatorial, priority-based, and frequently update to reflect improvements in standards of healthcare.

Here, I present a few options for pre-processing this important variable in both SQL and python for later use in machine learning models and predictive analytics, such as predicting whether a future claim will get denied.

CPT codes outline every procedure performed by US healthcare providers. …

Living at the intersection of healthcare and data science