Telecom industry generate huge volume of data every minute from different devices and infrastructures. Several thousand of alarm events are arriving in to network operation center of today's Ethio-telecom. Power and mobile failure alarm are the type of the alarm which is generated when the infrastructure is getting into trouble. To easily manage the faults by network management teams, the company needs to use machine learning to classify the severity to take appropriate action in order. Classical machine learning algorithm is very limited to process large volume of data like alarm history data used in this study. The data is taken from Ethio-telecom NNOC alarm history event. Big data analytics process model such as, problem definition, data source identification, data gathering, data preprocessing, analytics modeling and model evaluation are used to achieve the objective of the study. Scalable classification algorithms, logistic regression (SGD) and AdaptiveLogistic regression were utilized to build the classification model. To get the highly accurate model, 16 experiments are carried out by changing parameters using Apache mahout as a library. The built model is evaluated using confusion matrix to calculate: accuracy, recall, precision and the time taken to train the model. The most prominent model to classify the severity of the alarm history datasets of Ethio-telecom is AdaptiveLogistic regression with 20% of the data set used for testing with a classification accuracy of 100%. The result of this study indicates that applying scalable machine learning algorithms to classify large volume of alarm history event is very promising. So, it is better to use scalable machine learning library, apache mahout run on top of hadoop and independent of hadoop which supports different classification, clustering and collaborative filtering algorithms for very large datasets. In this study, all classification algorithms are not experimented and the selected algorithms are only executed in sequential mode. Therefore, future research directions are forwarded to utilize more algorithms and process in parallel using Mapreduce programing model.