Diabetes Simple Diagnosis

Repository created for validation of a Machine Learning model that predicts the diagnosis of diabetes in patients.

^{_{Image from Mohamed_hassan by Pixabay}}

Part 1

In the first stage, exploratory analysis is a general analysis of the dataset to understand all the information we have available in the database.

• Alteration in the name of the ID column.
• Validation of the types of variables in the data set.
• Validation if there are null values in the base.
• Validation of the dataset size, number of rows and columns.
• And also validation of the cardinality of the variables, especially the categorical variables.

Part 2

In the second stage, data pre-processing, the database was treated in such a way that it was as compatible as possible for Machine Learning models.The closer the dataset is to the low-level language, the better it will be interpreted by the model and the better the final result will be.

• Treatment of the Gender column, applying the get_dummies() method, which directly transforms categorical variables.
• Normalization of the columns 'Age','BMI','FBS','HbA1c_level' using the RobustScaler() method
• Definition of variables, target and explanatory, that will be used by the model.

Explanatory variables: ‘High_BP’, ‘Smoking’, ‘Female’, ‘Male’, ‘Other’, ‘Age_scaler’, ‘BMI_scaler’, FBS_scaler’, ‘HbA1c_level_scaler’.

Part 3

Model training stage, different models were trained using the Decision Tree, KNN and Logistic Regression algorithms. The performance metric used to measure the results of the models was accuracy.

You can check out more about accuracy and the main performance metrics for classification models in this article I wrote on Medium.

The results were:

• Decision Tree:0.9410
• KKN: 0.9550
• Logistic Regression: 0.9538

Part 4

The dataset is analyzed again to see if there is anything that can be changed to further improve the results. There is a small correlation between the Age variable and the target variable.Older patients are more likely to have a positive diabetes diagnosis than younger patients.

The Age_scaler column is replaced by the new column that is based on the patient's age group, dividing by classes from 0 to 4, the younger the age the closer it will be to zero, and the older the age the closer it will be to 4.

Explanatory variables: ‘High_BP’, ‘Smoking’, ‘Female’, ‘Male’, ‘Other’, ‘BMI_scaler’, FBS_scaler’, ‘HbA1c_level_scaler’, ‘Age_range’.

The results were:

• Decision Tree:0.9516
• KKN: 0.9523
• Logistic Regression: 0.9523

There was an improvement in the Decision Tree model but there was a worsening in the other two models.

Part 5

Again the dataset will be analyzed to see if we can identify another opportunity. The BMI (Body Mass Index) variable has a good correction with the target variable. Patients with BMI above 30 are considered diabetic.

The BMI_scaler variable is replaced by a new column that has a value of 1 for patients with a BMI above 30, and a value of 0 for values below 30.

Explanatory variables: ‘High_BP’, ‘Smoking’, ‘Female’, ‘Male’, ‘Other’, FBS_scaler’, ‘HbA1c_level_scaler’, ‘Age_range’,’BMI_range’.

The results were:

• Decision Tree:0.9619
• KKN: 0.9562
• Logistic Regression: 0.9516

The Decision tree model had an improvement of almost 1% in its results. And the other models maintained the same results as the first training, around 95%.

This last decision tree model was chosen to predict the test base. In this prediction, the test base had a result of 0.9643. A result of 96% and a result very similar to the result of the validation base.