A Comparative Analysis Of Machine Learning Algorithms For Subscription Fraud Detection The Case Of Ethio Telecom

In these days due to the development of affordable technologies, the numberrnof subscribers and revenue-generating increased over the past few years in therntelecommunication industry. However, advancements of the telecom industry providesrncertain appearances that stimulate fraudsters. One of the common and predominantrnfraud types is subscription fraud. It is usually the precursor to otherrnfraud types. Since 2013rnsubscription fraud is listed as a top-five predominantrnfraud type. Subscription fraud alone causes billions of dollar losses of telecommrncompanies.rnThis thesis is conducted on comparative performance of three supervised machinernlearning algorithms Artificial Neural Network (ANN), Support Vector Machinern(SVM) and J48rn, done using two classification techniques. Before analyzingrnand comparing the algorithms Call Detail Record (CDR) data were collected, relevantrnfeatures were selected and various preprocessing techniques such as featurernselection, data cleaning, shaping of data frame and feature types were performed.rnAs a result, J48rnalgorithm using Cross Validation (CV) options is found to be thernbest classifier algorithm by scoring 99rn.3rn% accuracy followed by the two algorithmsrnhighest scores of ANN ( CV ) and SVM (ST) with 97rn.51rn% and 96rn.0rn% respectively.rnThis result happens because of J48rn’s capable of learning disjunctive expressions inrnaddition to it reduced error pruning. Pruning decreases the complexity in the finalrnclassifier, so that improves predictive accuracy from the decrease of over fitting.

Get Full Work

Report copyright infringement or plagiarism