International Journal of Computer Applications (0975 – 8887) Volume 185 – No.13, June 2023 43 Predictive Framework for Advanced Customer Churn Prediction using Machine Learning Dinesh Kumar Jena Abhyarthana Bisoyi Electronics & Instrumentation Engg. Odisha University of Technology and Research Bhubaneswar, India Aruna Tripathy Electronics & Instrumentation Engg. Odisha University of Technology and Research Bhubaneswar, India ABSTRACT In recent years, the telecom sector has been burgeoning to satisfy the demand of mobile subscribers and telecom service providers. The increase in number of mobile subscribers and competition among providers, results in the creation of “churners”. These are the subscribers who tend to switch from the current telecom service to another. The detection of these churners is called “churn prediction”. This prediction has become a major challenge for telecom companies. The main purpose of customer churn prediction is to estimate the number of subscribers those who want to quit the current service provider by providing specific solutions to retain them. This paper proposes methods for the estimation of churners by applying different classification techniques and estimates the differences between them. The performance is measured by taking different parameters like accuracy, precision, recall, etc. In this paper, the various performance measurement and comparison are done by using the dataset collected from American Telecommunication Company. All the proposed work is based on Machine Learning, inculcating the supervised learning. In addition to all, a single test-bed is designed as a user interface to predict the individual customer according to different attributes. Keywords BigML, Churning, Customer churn prediction, Decision tree, Machine Learning, Random Forest, Supervised learning, Telecom, User Interface 1. INTRODUCTION Now-a-days the telecom industry is facing a competition among each other due to the entry of Reliance Jio into the market. In the middle of 2017-19, the loss incurred by service providers other than Jio, touched a high value which never occurred before. The onset of Jio created a competition among the service providers to retain their existing customers for a long-term profit [1-3]. Customer churn prediction plays a vital role to tackle the above said situation. Basically, customer churn refers to the inclination of existing customers of the current service providers towards the any other telecom service providers. The number of customers who leave the current provider in a particular interval of time are referred as churners. In the above scenario—to retain the customers and to keep a track of them—each telecom company should have a good predictive customer churn framework. Customer churn prediction is a prominent method to check the indirect feedback of customers as well as the profitability. Some statistics show the importance of churn prediction to the customer retain capacity of the company. One of the studies shows that, 1% increase in the customer retains campaign may result in 5% increase in the overall values of the company [5]. In the telecommunication industry, the monthly rate of customer churn is 2.2% and the annual rate of customer churn is 27% [4]. If we consider Europe and America, the yearly cost of customer churn is 4 billion dollars approximately. It is seen that if a company retain 1.5 million customers by increasing the rate at 1%, then this may add a benefit of 54 million dollars for the same year. Prodigious researches have been conducted for the creation of different models, development of new algorithms, development of existing models and classifiers, comparison of algorithms, efficiency, etc in order to retain the churners. In 2017, model was proposed by using decision tree classifier to solve unbalanced problem, scatter problem, high dimensional data in telecom data set [1]. In 2015, the rotation forest approach was followed to evaluate the prediction model, which is compared by C4.5 decision tree and ant-miner [2]. In addition, k-means clustering and down sampling to preprocess the data followed by the decision tree C5.0 and random forest for training and evaluation of data set has been reported in [3]. Further an attribute derivation process was implemented to increase the correct prediction rate [4]. Then, a Bayesian Belief network method was attempted in a study which was mentioned in [5]. Further the paper proposed support vector machine to increase the performance and an increased accuracy was obtained by using two different rules extraction method. Two methods like AntMiner+ and ALBA were implemented in [6-8]. A data mining model was developed for for the telecom churners about some churn prediction in the telecom sector in [9-10]. In addition, a hybrid model was developed as a combination of two artificial neural networks and a second hybrid model was a combination of self-organizing maps and artificial neural networks. [11-12]. In 2011, parameters such as F-measure and accuracy, which could be helpful in the development of Machine Learning algorithms were discussed in [13]. Various novel methods using Convolutional Neural Network (CNN) and other supervised learning methods for churn prediction have been reported in the recent time [14-20]. The recent work in churn prediction includes methods based on unstructured data [21-22]. Also, the decision tree has been implemented often in image processing-based application, as seen in [23-24]. These days random forest is being used in myriads of applications, as it provides better accuracy [25-28]. Supervised learning algorithms have been modified to self-supervised algorithms providing better results in image-based applications [29-33]. The major issues arising due to customer churning and its solution lies in determining the factors that leads to churning [34-35]. A predictive framework by decision tree and random forest approach is proposed in this paper after processing the data by random sampling method. And then, by taking the evaluation and result into account, a user interface is prepared to predict the individual churners according to different attributes.