Case study

Machine Learning Model for Future Diabetes Level Prediction

About Client

To determine the scope of the Hba1c level for diabetes, the Machine Learning model was applied with predictive analysis. It was performed with the aim to create future awareness and also reduce diabetes levels among individuals.

Problem STATEMENT

ML Model Building Process

Problem understanding
Statistical Analysis of data through Exploratory Data Analysis (EDA)
Data cleaning/wrangling
Model using ML Algorithms
Model evaluation/deployment

Exploratory Data Analysis

An overview of all predictor variables whether it is numerical, categorical, or dependent target variables.
Distributing data as per the target variables.
Finding and describing the shape, head, and information.
Analyzing the numerical and categorical features.
Listing and filling the features of missing values.
Transforming the log.
Building a relation of all features to the target variables.
Listing the numerical features and their coefficient correlation to the target.

Solution

Modeling

Applying the machine learning model to generate a pipeline for the loan amount prediction. The algorithms used:

Linear Regression
KNeighbors Regressor
Decision Tree Regressor
SVR or Support Vector Regressor
Random Forest Regressor
XGB Regressor

Evaluation

The metrics used for evaluating predictions on regression machine learning are:

Mean Absolute Error (MAE) – the sum of the absolute difference between predictions and actual values.
Mean Squared Error (MSE) – the mean absolute error that provides a gross idea of the magnitude of the error
R Squared or Root Mean Square Error – provides an indication of the goodness of a fitting set of predictions to their actual values.

Technical Architecture

Sample Dataset Feature Inputs

We considered the human behavior attributes like smoking status, BMI, height, weight, age, hemoglobin, etc that played a vital role in the prediction of the Hb1Ac:

Deep Learning Approach

Leveraging the state-of-the-art Deep Learning algorithm MiME (Multi-level Medical Embedding) we implemented the EHR data records. MiME is a novel EHR embedding algorithm that takes into account the connections between diagnoses and corresponding treatments and it has already shown promising results if data is provided appropriately. The embedding included different levels of information, like: