International Conference on Advances in Computing and Technology (ICACT)
Permanent URI for this communityhttp://repository.kln.ac.lk/handle/123456789/15607
Browse
4 results
Search Results
Item Mobile Telecommunication Customers Churn Prediction Model(3rd International Conference on Advances in Computing and Technology (ICACT ‒ 2018), Faculty of Computing and Technology, University of Kelaniya, Sri Lanka., 2018) Chathuranga, L. L. G.; Rathnayaka, R.M.T.B.; Arumawadu, H.I.The present Sri Lankan mobile industry is extremely dynamic, with new services, technologies, and carriers constantly altering the landscape. Then customers have more choices. So, Predict customer churn is one of the most challengeable target in the telecommunication industry today. The major aim of the study is develop a customer churn prediction model by considering some soft factors like monthly bill, billing complaints, promotions, hotline call time, arcade visit time, negative ratings sent, positive ratings sent, complaint resolve duration, total complaints, and coverage related complaints. This study introduces a Mobile Telecommunication customer churn prediction model using data mining techniques. In this study, three machine learning algorithms namely logistic regression, naive bayes and decision tree are used. Indeed, twenty attributes are mainly carried out to train these three algorithms. Furthermore, the back propagation neural network was trained to predict customer churn. Data set used in this study contains 3,334 subscribers, including 1,289 churners and 2,045 non-churners. According to the results, the trained neural network has two hidden layers with 25 total neurons. The proposed Artificial Neural Network result gives 96% accuracy for mobile telecommunication customer churn prediction. The estimated results suggested that the proposed algorithm gives high performances than traditional machine learning algorithm.Item Introducing Novel Classification Methodology to Detect Kidney Disease Patterns in Sri Lanka(3rd International Conference on Advances in Computing and Technology (ICACT ‒ 2018), Faculty of Computing and Technology, University of Kelaniya, Sri Lanka., 2018) Chathurangi, K.A.A.; Kapila, R.M.; Rathnayaka, T.The healthcare sector has vast amount of medical data which are not properly analyzed and mined to discover useful information and interesting patterns. Applying data mining techniques on such domain can help medical practitioners to predict even the crucial diseases with ease. This study introduced a novel kidney disease classification methodology in Sri Lankan domain using data mining techniques. Basically there are two types of kidney diseases that can be found in Sri Lanka namely Chronic Kidney Disease (CKD) and Acute Kidney Disease (AKD). The aim of this work is building a model to predict whether a person has a risk on having a kidney disease or not and a model for CKD prediction. The data collected from 108 patients are used to train and test the models. Random Forest algorithm and a multilayered feed forward neural network were used to build the models. Result of this study is a modified Artificial Neural Network with 2 hidden layers to detect kidney disease which gives 0.80952 accuracy and a model with the combination of Random Forest algorithm and Artificial Neural Network with 3 hidden layers for CKD prediction which gives 0.81395 accuracy for testing data. The constructed models give high accuracy and minimum error rate when comparing with the other data mining algorithms.Item A Simple Machine Learning Approach for Identifying Promotional Short Message Service (SMS) Messages.(Faculty of Computing and Technology, University of Kelaniya, Sri Lanka., 2017) Dias, D.S.; Dias, N.G.J.Mobile phones play an integral part in the modern lives of humans. Short Message Services (SMS) Messages have become a popular mode for simple communication. Its’ simplicity, costeffectiveness and large audience has attracted the attention of advertising industry to send targeted promotional messages to mobile phones. In Sri Lanka, a survey conducted in Colombo, yielded that 3 out of 5 SMS messages received our promotional messages. Even though extensive research has been carried out in detecting junk SMS messages, the amount of research conducted on filtering promotional SMS messages is rare. The purpose of this research is to evaluate the success and accuracy of utilizing a simple machine learning algorithm to identify promotional SMS messages. Here, we have used a feed-forward neural network based on a statistical model, which was trained with a training data set consisting of promotional as well as non-promotional messages. Each test message was broken down in to individual words and filtered through by cleaning to form keywords which will have consist of a weight and probability value. With each message that is used to train, these values will be updated according to whether it is a promotional or a non-promotional message. When a message is tested through this neural network, the words of the message will be matched against the keyword’s weight and probability, which will then calculate a resultant probability. By setting a par-value, we can classify the test as a promotional or a non-promotional message. The proposed model yielded a 100% accuracy when tested within the given test data set. In order to get successful results for broader test data sets, the model has to be trained comprehensively with proper amount of promotional and non-promotional messages. Optionally, the results obtained from the feed forward neural network for incoming messages, can then be fed back in to the feed forward neural network for further training. As future work, we intend to take this solution to an android-based mobile application that extracts promotional messages from the incoming SMS messages as well as from a server, and display them to the user based on his preferences.Item Data mining approach for Sales Prediction(Faculty of Computing and Technology, University of Kelaniya, Sri Lanka, 2016) Rathnadiwakara, A.S.K.; Liyanage, S.R.Nowadays predictive analysis is more popular among companies to improve their business profits. Those companies differentiate what they do from “Data Mining”. The characteristic deduction is that data mining is limited to the discovery of patterns, whereas predictive analytics allows the application of the patterns to new data to predict unknown values. The main aim of data mining is to extract knowledge from the data at hand, increasing its intrinsic value and making the data useful. Today most business areas using many strategies to improve their business profits. They are mostly use traditional methods. Therefore, the company’s efficiency and profitability, goes to the critical situation. When considering today’s business arena, the most important things are good efficiency and correct strategies for a business. Converting to the new technologies companies can achieve their business goals and they can reveal their sales life-cycle. This research proposes for a medium scale tyre dealing company which is situated in Colombo. It is important to company to accurately predict their future order details and salary income before having unprofitable occasion. Company could conduct sufficient stock when talking prediction support. That is the best solution to reduce time, save importing cost, growth for income and manage resources. Data mining algorithms and techniques used for the prediction process and used MS SQL Server 2008 R2 with Analysis Server and Business Intelligent Development Studio for modeling process. Analysis Services contains number of standard data mining algorithms. Decision Tree, Neural Network and Clustering data mining models were attempted for the prediction. Decision Tree is a graph of decisions and their possible consequences, represented in form of branches and nodes. A Neural Network is a parallel distributed processor that has a propensity for experiential knowledge and making it available for users. Clustering is used to place data elements into related groups without advance knowledge of the group definitions. The best algorithm was selected for each model and it focused on five main attributes which were referred to as factors affecting a sales process such as Item Code, Item Type, Item Quantity, Item Value, Item Sold Date, etc. variables were used in data mining process. Among those variables five variables were selected for the mining process. Dataset arranged with 30% data for testing process and 70% data for the training process. According to the predicting probabilities, Decision Tree algorithm were performed 99.53%, Neural Network algorithm were performed 73.36% and Clustering algorithm were performed 67.79%. Clustering model belongs to the lowest predict probability value. Therefore Clustering model was the worst model. Decision Trees model contains highest predicted value 99.53%. Therefore it was the best model. Neural Network model was also a good model. The Score results indicate that Decision Trees mining model has the best score 1.00 and followed by Neural Network mining algorithm with score of 0.92 and clustering mining algorithm with 0.94. Considering the data mining lift chart for mining structures, it graphically represents the improvement that a mining model provides when compared against a random guess, and measures the change in terms of a lift score. By comparing the lift scores for various portions of the dataset and for different models. According to the Lift chart representation, Decision Trees curve present in upper in the chart with comparing other carvers. By considering lift chart, score and target population with predicting probabilities, Decision Trees algorithm was the best one for prediction process. Finally, Data mining model was implemented using Decision Trees algorithm. According to these predicting results, the company can handle their imports optimizing the available resources; storage, time, money. Therefore this research would benefit the Company to improve their incomes.