Diabetes-Prediction-Using-Machine-Learning-Algorithms

Diabetes is a prevalent chronic condition that can seriously endanger human health. When blood glucose levels are greater than normal, diabetes can be diagnosed whether is caused from various kinds of biological factors. Machine learning algorithms can be used to predict diabetes by selecting and processing relevant medical data, training, and validating the model, and incorporating it into clinical decision support systems.

Model Accuracy

	After Feature Selection	After Hyperparameter Tuning	Final Accuracy (Test Set)
Logistic Regression	79.60 %	79.60 %	78.04 %
K-Nearest Neighbors	90.80 %	91.33 %	88.76 %
Decision Tree	89.60 %	91.60 %	90.04 %
Support Vector Machine	83.20 %	85.97 %	91.24 %
Random Forest	94.13 %	94.40 %	94.20 %
XGBoost	95.87 %	96.40 %	96.52 %
Light-GBM	94.93 %	95.60 %	96.64 %

Precision-Recall Curve

Confusion Matrix & Classification Report

Logistics Regression

	Predicted Negative	Predicted Positive
Actual Negative	1464	198
Actual Positive	351	487

Performance Measures	Precision	Recall	F1-Score	Support
0	0.8066	0.8809	0.8421	1662
1	0.7109	0.5811	0.6395	838
Accuracy	0.7804	0.7804	0.7804	0.7804
Macro Avg	0.7588	0.7310	0.7408	2500
Weighted Avg	0.7745	0.7804	0.7742	2500

K-Nearest Neighbors

	Predicted Negative	Predicted Positive
Actual Negative	1578	84
Actual Positive	164	674

Performance Measures	Precision	Recall	F1-Score	Support
0	0.9059	0.9495	0.9271	1662
1	0.8892	0.8043	0.8446	838
Accuracy	0.9008	0.9008	0.9008	0.9008
Macro Avg	0.8975	0.8769	0.8859	2500
Weighted Avg	0.9003	0.9008	0.8995	2500

Decision Tree

	Predicted Negative	Predicted Positive
Actual Negative	1572	90
Actual Positive	129	709

Performance Measures	Precision	Recall	F1-Score	Support
0	0.9242	0.9458	0.9349	1662
1	0.8874	0.8461	0.86626	838
Accuracy	0.9124	0.9124	0.9124	0.9124
Macro Avg	0.9058	0.8960	0.9005	2500
Weighted Avg	0.9118	0.9124	0.91195	2500

Support Vector Machine

	Predicted Negative	Predicted Positive
Actual Negative	1546	116
Actual Positive	165	673

Performance Measures	Precision	Recall	F1-Score	Support
0	0.9036	0.9302	0.9167	1662
1	0.8530	0.8031	0.8273	838
Accuracy	0.8876	0.8876	0.8876	0.8876
Macro Avg	0.8783	0.8667	0.8720	2500
Weighted Avg	0.8866	0.8876	0.8867	2500

Random Forest

	Predicted Negative	Predicted Positive
Actual Negative	1547	56
Actual Positive	107	790

Performance Measures	Precision	Recall	F1-Score	Support
0	0.9353	0.9651	0.9500	1662
1	0.9338	0.8807	0.9065	838
Accuracy	0.9348	0.9348	0.9348	0.9348
Macro Avg	0.9346	0.9229	0.9282	2500
Weighted Avg	0.9348	0.9348	0.9344	2500

XGBoost

	Predicted Negative	Predicted Positive
Actual Negative	1622	40
Actual Positive	47	791

Performance Measures	Precision	Recall	F1-Score	Support
0	0.9718	0.9759	0.9739	1662
1	0.9519	0.9439	0.9479	838
Accuracy	0.9652	0.9652	0.9652	0.9652
Macro Avg	0.9619	0.9599	0.9609	2500
Weighted Avg	0.9651	0.9652	0.9652	2500

Light-GBM

	Predicted Negative	Predicted Positive
Actual Negative	1629	33
Actual Positive	51	787

Performance Measures	Precision	Recall	F1-Score	Support
0	0.9696	0.9801	0.9749	1662
1	0.9598	0.9391	0.9493	838
Accuracy	0.9664	0.9664	0.9664	0.9664
Macro Avg	0.9647	0.9596	0.9621	2500
Weighted Avg	0.9663	0.9664	0.9663	2500

Model Deployment

Methodology

Overview Procedure from Data Collection to Model Deployment:

Data Visualization

Data visualisation is a useful tool that enables people to grasp and comprehend complex data more easily. It entails presenting data in a graphical or visual style, which can aid in identifying patterns, trends, and relationships that may not be obvious when looking at raw data.

Skewness of Attribute Distribution

Histogram for each attribute in dataset

Bar Graph for Distribution of Diabetic

Box and Whisker plot

Data Preprocessing

Data preprocessing is a method that is used to enhance the data's quality prior to apply mining and ensuring that the data will produce high quality mining results. Preprocessing of data includes:

• Data Cleaning
• Data Integrating
• Data Transforming
• Reducing Data

Pearson Correlation

Pearson correlation is a statistical metric used to quantify the linear relationship between two continuous variables. The Pearson correlation coefficient is a number between -1 and 1, with 1 indicating a perfect positive linear link, -1 indicating a perfect negative linear relationship, and 0 indicating no linear relationship.

Correlation Matrix for each feature

Correlation Matrix between features and outcome

From the correlation matrix above, we can observe the correlation coefficient of ‘PatientID’ is smaller. It can be considered to remove it from datasets if we want to save the time consume and save the computation cost during model training.

StandardScaler

StandardScaler is a preprocessing tool for standardising the features in a dataset. This is important because most of the machine learning algorithms assume that the input data is normally distributed and that all characteristics are of equal magnitude. StandardScaler normally will be applied logistics regression and support vector machine, PCA and so on during training process.

StandardScaler Data

Model Training With Wrapper Method and Hyperparameters Tuning

Wrapper method is one of the feature selection techniques that is a strategy for selecting the most important features to include in a model. This strategy entails evaluating several features combinations and selecting the one that results in the best model performance.

Hyperparameter tuning is the process of determining the best settings for machine learning. Hyperparameters that are not learned from data but can be adjusted or used as defaults before training the model. The model can be optimised by tuning the hyperparameters to achieve the best feasible performance on a given dataset.

Weightage of Training, Validiation and Test Set

Logistic Regression

Feature Selection with SFS:

Hyperparameters tuning in Logistic Regression with Grid Search:

Hyperparameter	Values
max_iter	10000
C	0.059
penalty	l2
solver	lbfgs

K-Nearest Neighbors

Feature Selection with SFS:

Hyperparameters tuning in K Nearest Neighbors with graph method:

Decision Tree

Feature Selection with SFS:

Hyperparameters tuning in Decision Tree with Grid Search:

Hyperparameter	Values
max_depth	10
max_leaf_nodes	40

Support Vector Machine

Feature Selection with SFS:

Hyperparameters tuning in Support Vector Machine with Random Search:

Hyperparameter	Values
max_depth	40
max_features	auto
n_estimators	100
min_samples_leaf	1
min_samples_split	1

Random Forests

Feature Selection with SFS:

Hyperparameters tuning in Random Forests with Grid Search:

Hyperparameter	Values
gamma	0.0001
C	1000
kernel	rbf
probability	True

XGBoost

Feature Selection with SFS:

Hyperparameters tuning in XGBoost with Random Search:

Hyperparameter	Values
max_depth	40
subsample	0.7
min_child_weight	1
learning_rate	0.2
gamma	0.2
colsample_bytree	0.2

Light-GBM

Feature Selection with SFS:

Hyperparameters tuning in Light-GBM with Random Search:

Hyperparameter	Values
max_depth	15
num_leaves	30
n_estimators	100
subsample	0.8
colsample_bytree	0.1