Proper

Abstract

Ensemble method is a meta-algorithm to build strong classifiers based on a set of weak classifiers. This work explores some ensemble classifiers on UCI Arrhythmia Dataset to classify the heartbeat records. Each record corresponds to a person, and is the features extracted from the person’s raw ECG data over a period. Two popular ensemble classifiers XGBoost and RandomForest are used, and the classic LogisticRegression is tried as a comparison. These classifiers are applied on all the 279 features in the dataset, and predict the heartbeat categories for the records. XGBoost and RandomForest perform better than even well-tuned LogisticRegression. We build VotingClassifier based on the ensemble voting meta-algorithm and the above three built-in classifiers, and it outperforms even well-tuned XGBoost and RandomForest. The best prediction accuracy 76% is achieved by the VotingClassifier in this multiclass classification problem. This result is comparable to many other findings that uses similar or different classifiers.