PIMA-Diabetes_Prediction_ML

This Notebook provides you a deep insight on Machine Learning Modelling and HyperParameter Tuning.

Prediction whether a women has diabetes.

Topics

Data Analysis
Data Scaling
Data Modeling
Hyper-parameter Tuning.
Cross Validation
Predictions

Data Analysis:

The Dataset has 768 rows and 9 columns
Target Columns is: ‘Outcome’
We plot graphs to see the variation of each Independent variable

We plot graph to verify the outliers so that we can clean our dataset.

Plot to check the correlation between the data points.

Data Scaling:

RobustScaler is a transformation technique that removes the median and scales the data according to the quantile range (defaults to IQR: Interquartile Range). The IQR is the range between the 1st quartile (25th quantile) and the 3rd quartile (75th quantile). It is also robust to outliers, which makes it ideal for data where there are too many outliers that will drastically reduce the number of training data.

Data Modeling

We use the state-of-the-art XGBClassifier to predict the outcome.(whether a person has diabetes)
Use GridSearchCV for hyperparameter tuning on
- learning_rate
- n_estimators
- max_depth
- min_child_weight
- reg_alpha
- reg_lambda

We get a accuracy score of 82.1429%

Graph plot for ROC and AUC

We also use Logistic Regression with cross-validation technique to check:
- log_loss
- accuracy
- f1 score
  
  We get a mean accuracy of 78.02%