There are four ECG arrhythmia datasets in here, each employing 2-lead ECG features. Datasets obtained from PhysioNet are MIT-BIH Supraventricular Arrhythmia Database, MIT-BIH Arrhythmia Database, St Petersburg INCART 12-lead Arrhythmia Database, and Sudden Cardiac Death Holter Database.

In each of the datasets, the first column, named "record" is the name of the subject/patient.

import pickle



import numpy as np

import pandas as pd

import seaborn as sns

import matplotlib

import matplotlib.pyplot as plt



plt.rcParams.update({'figure.figsize': (12.0, 8.0)})

plt.rcParams.update({'font.size': 14})



#default theme

sns.set(context='notebook', style='darkgrid', palette='colorblind', font='sans-serif', font_scale=1, rc=None)

matplotlib.rcParams['figure.figsize'] =[8,8]

matplotlib.rcParams.update({'font.size': 15})

matplotlib.rcParams['font.family'] = 'sans-serif'​

Read Dataset and check first 5 rows of the dataset

# Reading MIT-BIH Arrhythmia Dataset as an example

data_df = pd.read_csv('MIT-BIH Arrhythmia Database.csv', low_memory=False)



print("Number of rows in data =",data_df.shape[0])

print("Number of columns in data =",data_df.shape[1])

print("\n")

print("**Sample data:**")

print(data_df.head())

Description Of The Dataset

print(data_df.describe())

Checking for Null values

print(data_df.isnull().sum())

​

 Split the data into features and class labels

x_data = data_df.drop(['type','record'], axis = 1)

y_label = data_df[['type']]

print(y_label.value_counts())

Replacing Multiple Target Variable into Binary Class

# Transform multi-class labels into binary-class (arrhythmia and normal)

y_label.replace(['VEB','SVEB','F','Q'], 'arrhythmia', inplace=True)

y_label.replace(['N'], 'normal', inplace=True)

data_df['type'] = data_df['type'].replace(['normal','arrhythmia'],[0,1], inplace=True)

 Train-test Split

from sklearn.model_selection import train_test_split

X_train, X_test, y_train, y_test = train_test_split(x_data, y_label, random_state=101)

Feature Scaling

from sklearn.preprocessing import MinMaxScaler



min_max_scaler = MinMaxScaler()

X_train = min_max_scaler.fit_transform(X_train)

X_test = min_max_scaler.transform(X_test)



print(min_max_scaler.scale_)



pickle.dump(min_max_scaler, open('min_max_scaler.pkl','wb'))

importances = model.feature_importances_

# Sort the feature importance in descending order

sorted_indices = np.argsort(importances)[::-1]



feat_labels = data_df.columns[2:]



for f in range(X_train.shape[1]):

    print("%2d) %-*s %f" % (f + 1, 30,

                            feat_labels[sorted_indices[f]],

                            importances[sorted_indices[f]]))

Model training

from sklearn.ensemble import RandomForestClassifier

model = RandomForestClassifier(random_state=101, n_estimators=150)

model.fit(X_train, y_train.values.ravel())

Model testing
a)Accuracy of the model

from sklearn import metrics

y_pred = model.predict(X_test)

print("Accuracy:",metrics.accuracy_score(y_test, y_pred))

b)Confusion Matrix

print("*** Confusion Matrix ***")

print(metrics.confusion_matrix(y_test, y_pred))

c)Classification Report

print("*** Classificstion Report ***")

print(metrics.classification_report(y_test,y_pred))