Atrial repolarization (Ta wave) characteristics has not been well studied for atrial arrhythmia diagnosis unlike other ECG waves, as it is obscured by QRS complex. Hence, this study proposes a new method to interpolate the Ta wave and aims to introduce Ta wave features as a predictor of atrial arrhythmias. ECGs from ten Sinus Rhythm (SR), ten Sinus Tachycardia (ST) were recorded using EDAN SE-1010 ECG system, and ten Atrial Tachycardia (AT) ECGs were extracted from the public database. Lead-II 10-second ECGs were pre-processed in MATLAB R2021a version, and spline interpolation model was employed to visualize the complete Ta wave within QRS complex, which was otherwise hidden. Further, Ta wave feature sets were prepared separately and analyzed for all three rhythms. Through statistically significant features, Machine Learning (ML) classifiers were implemented for optimal model selection. Ta peak duration and amplitude, Ta duration/amplitude and Ta area are the most potent feature in heart rhythm classification. The extra-trees model possessed the highest accuracy values of 94.31 % with the computational time of 0.64 seconds using Ta wave features. The findings of this study indicate that Ta wave features have the potential to assist clinicians in improved detection of atrial arrhythmias. Keywords: Atrial repolarization, Atrial tachycardia, Classification, Electrocardiogram, Machine learning, Spline interpolation ALL ARE CORDIALLY INVITED