Heart Disease Prediction

Simple project to determine the degree of narrowing blood vessel walls based on the features of the patient

Tools used:

Language: Python
Libraries: Numpy, Pandas, Seaborn
Frameworks: Tensorflow, Keras, Scikit-Learn, Flask

The dataset is taken from UCI
Features:

age: age in years
sex: sex (1 = male; 0 = female)
cp: chest pain type (1 = typical angina; 2 = atypical angina; 3 = non-anginal pain; 4 = asymptomatic)
trestbps: resting blood pressure (in mm Hg on admission to the hospital)
chol: serum cholestoral in mg/dl
fbs: fasting blood sugar > 120 mg/dl (1 = true; 0 = false)
restecg: resting electrocardiographic results (0 = normal; 1 = having ST-T wave abnormality; 2 = showing probable or definite left ventricular hypertrophy by Estes' criteria)
thalach: maximum heart rate achieved
exang: exercise induced angina (1 = yes; 0 = no)
oldpeak: ST depression induced by exercise relative to rest
slope: the slope of the peak exercise ST segment (1 = upsloping; 2 = flat; 3 = downsloping)
ca: number of major vessels (0-3) colored by flourosopy
thal: (3 = normal; 6 = fixed defect; 7 = reversable defect)
target: diagnosis of heart disease (angiographic disease status) (0 = < 50% diameter narrowing; 1 = > 50% diameter narrowing)

SOURCE
The dataset has 13 features. The target variable is the last one
There are 1025 entities in the dataset, each with their own feature values. Each entity represents a patient
The dataset is already cleaned, with no null values
Using the pandas correlation matrix and seaborn heatmap, we see:
- oldpeak and exang are negatively correlated
- cp and thalach are positively correlated
From the above observation, it becomes clear that:
- The chance of diameter narrowing reduces with exercise
- Higher heart rate and chest pain indicates diameter narrowing
Exercising regularly and maintaining a healthy heart rate can clearly reduce the chances of diameter narrowing

SOURCE
We drop the target variable from the dataset and use the remaining features to predict the target variable
We will use 80% of the dataset for training the neural network
The model we used is a simple sequential model, consisting of only 1 hidden layer
The structure of the model is as follows:
- Input layer: 64 neurons, ReLU activation
- Hidden layer: 32 neurons, ReLU activation
- Output layer: 1 neuron, Sigmoid activation
Since the target is binary, we have used Sigmoid as the activation function of the output layer, and the optimizer is Adam
Since the dataset is relatively small, we have created this simple NN structure and also are dropping 10% training data while leaving each layer. All this has been done to avoid overfitting
Training accuracy is 83%
Validation accuracy is 78%
I have tried tweaking the hyper-parameters multiple times, but this seems to be the most optimal. I did not increase the complexity of the model just for the sake of improving it's accuracy, since it will come at a cost of overfitting
The model is saved in the file heart_model.h5
As shown in the notebook, we can load the model and use it to predict the target variable for any new entity
The model throws result in floating point numbers, while the target should be integer (binary)
We converted the output to integer and then tested against an unseen dataset. The accuracy was 78%
The model seems to have been properly trained