Efficient Prediction of Internet Service Provider using Machine Learning

Efficient Prediction of Internet Service Provider using Machine Learning

Vol. 12 Issue 07, July-2023

Efficient prediction of internet service provider using machine learning

Sanju Patil

School of ECE

KLE Technological University

Hubli, INDIA

Suneeta V. Budihal

School of ECE

KLE Technological University

Hubli, INDIA

Saroja V. Siddamal

School of ECE

KLE Technological University

Hubli, INDIA

AbstractIn the new trend of process of improving the network services there are lot of challenges faced in commu- nication domain among which selection of best internet network is also a major one. To overcome this we designed a machine learning model which predicts the best service provider among given data and helps us to improve the service by providing a feedback analysis on the predicted information based on several characteristics. Machine learning algorithms are being used here as it can provide a broader class of several alternative analysis methods which are best suited to modern data set.

1. INTRODUCTION

   We are currently contributing to the definition of what 5G networks will resemble. Multiple work items that will result in the definition of a novel 5G radio and architecture have already been started by the 3GPP [1].Numerous white papers detailing the leading vendors perspectives on 5G networks and archi- tectures are being published. In order to fund research for 5G networks, the EU commission has established a significant 5G Infrastructure Public Private Partnership (5GPPP) programme [2]. It is evident from all of these convergent perspectives that network management in future 5G networks will have to deal with a completely new set of issues.

   In this situation, it is already commonly acknowledged that new protocols must be set up in order for the network to become more intelligent, self-aware, and self-adaptive. Since Release 8, the Self-Organising Network (SON) has been a part of 4G LTE networks, which is the first step in this approach. However, given the enormous complexity of these networks, this idea needs to be refined further for 5G. As we previously noted in, control and management functions generate a significant amount of data during routine operations in 4G, and more data is gathered in 5G as a result of the densification process, heterogeneity in layers and technologies, the added complexity of control and management in NFV and SDN architectures, and the growing importance of M2M and IoT communications.

   This paper proposes to predict the best internet service provider using machine learning algorithms. The paper in- volves a series of steps carried out, starting with understanding the problem definition and reviewing the data set, setting an end goal for a desired problem and carried out by listing the alternate solutions and selecting the best suited solution. Searching for the respective algorithms and understanding the

   working of algorithms to propose a solution for a our desired problem statement and then implementing the solution by testing and training the model and evaluating the accuracy and deploying the model.

   Every machine learning project is carried out through these steps: Although the identified solution is suitable for the prob- lem statement, but trying different models is required in order to know the variations in accuracy to build a proper model with best prediction analysis. This paper aims at improving the existing algorithms with some modifications so as to make the algorithms free of errors. Hence, it is essential for securing, stabilizing and increasing the efficiency. As virtualization is a great concern in this era, there is a huge demand for the good network speed in the society. This paper helps the service provider to implant the tower in required area as well as help the customer to choose the best network provider in their respective areas.

   1. Literature survey

      Authors in paper [3] have proposed the concept of mobile network tool based on prediction of data analytic, where they mainly focused on best network planning. It gives a proper definition to the service provided to the users and the resources used by the customers. This framework worked mainly in two steps. First, they proposed a model of service through the analysis of data collected from the networks in the form of different measures. Second, they changed the parameters and analysed the impact on QoS based on the previous learning. In this way, the performance is optimised to meet the targets focusing on Random Forest algorithms. Authors in paper[4] have given brief overview of Random Forest Algorithm. They have also discussed the key feature of Random Forest Algo- rithm i.e., node size, the total number of trees and features sampled, etc. In general they have highlighted the importance of Random Forest algorithm i.e., proof-based and vote-based. Authors in paper [5] have given a brief overview on feature extraction method that is based on deep belief networks and random forest algorithm; This algorithm is based on multi layer neural network to minimise the dimension of the provided data, and then, Random forest algorithm is applied.

      The algorithmic model consists of the following:

      • Data acquisition using wavelet denoizing

      • Feature extraction using deep neural network method

        Fig. 1: Flow diagram for the proposed methodology

      • Signal recognition using classification method.

        After collection of data and preprocessing the data, DNN is used for extraction of feature vectors from training and testing dataset. Then RF model is used for classification of training dataset and use testing dataset for validation purpose. Authors in paper [6] have given the importance to the variables that are important in predicting bovine viral diarrhea virus. The method in random forest algorithm has properties which makes it appealing for classification problems.

        Authors in paper [7] have proposed transmission rates and other physical characteristics of the network that are measured and improved and to some extent analysis of the network [8-10]. Traffic load is very important for optimisation of the proposed model and to control the services provided by the network. These services are important for control mechanism on network and also maintaining the maximum utilization of the service. Resource allocation [11-12] is optimised in such a way that it imitates with QoS .

   2. Problem Statement

      To Predict the best Network provider/ best Internet service provider across India in different states according to their network performance. This is a data set of government of India collected by TRAI using Myspeed application. The data is sampled from 1.3 million devices on which Network speeds were measured by Myspeed application of TRAI. The samples are taken from all the states of India, from various Service Provider. Characteristics present in the dataset are Service Provider, Technology Test type, Data speed(mbps), Signal strength and LSA. The dataset contains roughly equal

      no of download and upload tests, collecVteodl. f1r2omIssMuea0r7c,hJu2l0y1-28023 from various states of India and consisting of various service providers. Apart from the data sets for 4G and 3G network, the data also consists of signal strength while the speeds were measured.

   3. Objectives

      • Develop a Prediction based model.

      • Implement Random Forest algorithms using Machine learning.

      • Training and Testing of implemented model using ma- chine learning.

      • Deployment nd performance check of Random Forest algorithm using different.

        The above propose model helps in various different ways which acts an application for society to improve their services

      • Customers can Easily choose the best service provider by analyzing the nature of predicted service provider the customers can choose the best network for their area.

      • Improving the services by service provider by the pre- diction data available as a result of prediction the service providers can study the data and can improve the service for particular area or state as required.

2. PROPOSED SYSTEM FRAMEWORK

   The Figure 2 represents the System design of proposed model.

   Fig. 2: Architecture of System Design

   The given problem in this case as indicated will be solved by using Machine Learning model.specifically Random Forest implemented using python programming language with the help of multiple inbuilt libraries. Hence, as described in the

   Fig. 3: Flowchart of Random Forest

   following section we discuss the details of system design of architecture capable to solve the given problem and the same using python and libraries for the purpose.

   1. Analysis of Proposed model

      Supervised learning consists one of the algorithm named Random Forest Algorithm which is usually used for classifica- tion problems rather than regression problems. Random Forest algorithm is made up of multiple decision trees, as number of trees increases the more is the robust forest. Random Forest creates decision tree and predicts the output for each tree, then using the voting method it takes the best solution among all the trees. This method is considered because it is better to take output from multiple trees rather than selecting the solution from single decision tree. It increase the accuracy also overcomes over-fitting problem.

   2. Flow Chart

      Figure 3 shows the flow of Random Forest Algorithm. The data set is split into two segments training data (70%) and testing data (30%). Then N number of samples are taken from training data set to train the model. Then voting is done among all the decision trees created from those training samples. Then the decision tree with more numbers of votes is selected as the best prediction method.

   3. Functional Block Diagram

      Figure 4 is the functional block diagram of Random Forest The working of Random Forest Algorithm as sequence dia- gram of the proposed model is shown in Figure 4. Initially there is a group of decision trees created and voting is done among them by then the training and testing data is split then variables are chosen and stop conditions are applied for each chosen variable at next splits.then sorting of variables is done then index is calculated at each split and prediction error is calculated.

      Vol. 12 Issue 07, July-2023

      Fig. 4: Functional Block Diagram of Random Forest

      • Considering the training samples to be N and testing data samples to be M

      • m number of inputs is used to determine the decision at node of the tree where M should be greater than m

      • Choosing training samples for the tree

      • For each node of tree randomly m variables are chosen.

      • Taking results from each tree and best solution is chosen by voting method.

        For prediction a new sample is pushed down the tree. It is assigned the label of the training sample in the terminal node it ends up in. This procedure is iterated over all trees in the ensemble, and the average vote of all trees is reported as random forest prediction.

   4. Advantages of Proposed Model

      This model runs efficiently on large data set.

      • One of the advantage is that it handle thousands of input variables without variable deletion.

      • The model can provide estimates of what variables are important in the classification.

      • This model offers an experimental method for variable detection.

3. IMPLEMENTATION DETAILS

   In this section the details of requirements of the system are given. The implemented paper is capable of predicting the best Internet service provider across India in different states according to their network performance.

   1. System Requirements

      The system should be able to implement Machine learning Algorithm

      • The system should be able to implement Random Forest algorithms.

      • The system should be able to implement the SVM algo- rithm

      • The system should detect and Predict the result with maximum accuracy. The non-functional requirements are:

      • The system should be able to display plots that are the outputs of running code cells

      • The system should reduce the resource consumption by considerable amount.

   2. Software Requirements

      • Windows/Mac

      • Jupyter Notebooks/Google colab

      • Python 3.6 with Keras, Matplotlib, Numpy, Pandas, sklearn Libraries.

   3. Hardware requirements

      • Minimum 8GB RAM

      • 512GB hard disk

      • Minimum core i3 processor

   4. Random Forest Algorithm

      These steps represent the working of algorithm in creating the Random Forest model:

      Step1: Begin

      Step2: Collecting the data from TRAI

      Step3: Preprocessing of data using different functions such as concat, drop, getdummies

      Step3: Providing certain weightage for training and testing data

      Step4: Training the dataset with random frorest algorithm Step 5: Testing the results on testing dataset where accuracy is determined

      Step 6: Comparing the Random forest algorithm with SVM and Neural networks to differentiate the accuracy of each algorithm with respect to random forest algorithm

      Step 7: End

   5. Optimization

   Optimisation is a technique of finding set of inputs which leads to maximum accuracy and minimum function evaluation hence performing better. Many machine learning algorithms face problem from fitting the logistic regression to training ANN. The purpose of optimisation is to get the best design comparing to set of constraints or criteria. These include strength, longevity, productivity, reliability, utilization and effi- ciency. Optimisation is an important tool for making decisions and analysing the physical system. Optimisation is also defined as finding best solution from all the feasible solutions. Program or software optimisation is defined as process of modification

   No.	Algorithm	Training Accuracy	Testing accuracy
   1	SVM	87.29% Vo	l. 12 80.5%
   2	Random Forest	89.7%	87.5%

   Issue 07, July-2023

   TABLE I: Comparison of algorithm accuracy

   of software system to work more efficiently with minimal resources.

   Optimisation technique used in this paper is hyperparameter tuning. Hyperparameters refers to the parameters which helps in defining the model architecture. Some of the hyperparam- eters are constraints, learning rate. Hence searching of ideal hyperparameters which play very important role in increasing the efficiency and accuracy of the proposed model.

4. RESULTS AND DISCUSSIONS

   The following outputs were obtained when trained the model with Random Forest, SVM and Neural Networks Al- gorithms.

   Fig. 5: Random Forest Accuracy

   Fig. 6: Sample data set used for training model

   The trained model has an accuracy of 87.5 % with a prediction method implemented with RF Algorithm. When

   tried with different model comparing SVM with Random Forest for results analyzes. We plotted a confusion matrix for SVM model and also got an accuracy of 80.5 %.

   Fig. 7: Multiple Accuracy of the test cases for RF Model

   Fig. 8: Output plot over maximum depth and accuracy of RF Model

   Fig. 9: Correct Prediction of test case for RF Model

   Vol. 12 Issue 07, July-2023

   Fig. 10: False Prediction of test case for RF Model

   Fig. 11: Conclusion matrix for SVM Model

   So Random Forest was having more accuracy then that of SVM model. When we tried comparing Random Forest with Neural Network we plotted graphs between epoch vs loss and epoch vs accuracy and the accuracy was less compared to that of Random Forest so we choose Random Forest as model of implementation. The Table 1 represents the comparison of accuracy of models.

5. CONCLUSION

In Machine Learning for an algorithm with sufficient re- sources, it is important to get a fair chance of prediction. So far in this paper we focused on how to use machine learning algorithms and tools, By developing a prediction based model on best network service provider. Initially the dataset was taken from TRIA and some preprocessing of the dataset was carried out then an application of random forest algorithm was implemented where random forest regressor algorithm was applied. Which leads an accuracy of 87.5 % others

Fig. 12: Accuracy for SVM Model

Fig. 13: Plot of Model Accuracy vs Epoch for CNN Model

algorithms like SVM and Neural Networks were also tried for a defined problem statement. Machine learning is one of the leading carrier choices at present where projects like these can be implemented to create a well and easy environment for Telecommunication industries to obtain the prediction based data of best and worst network service provider in order to improve their services an provide a better service.

REFERENCES

[1] 3GPP on track to 5G, http://www.3gpp.org/ news-events/3gpp- news/1787-ontrack 5g.

[2] 5GPPP.The 5G Infrastructure Public Private Partnership, https:// 5g- ppp.eu/.

[3] Lorenza Giupponi and Josep Mangues-Bafalluy, A Mobile Net- work Planning Tool Based on Data Analytics, Volume 2017 https://www.hindawi.com/journals/misy/2017/6740585/

[4] Ke Li, Nan Yu, Pengfei Li, Shimin Song, Yalei Wu, Yang Li, Meng Liu, Multi-label spacecraft electrical signal classification method based on DBN and random forest,May 2017 https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0176614

[5] Serkan Balli, Ensar Arif segbas, Musa PerkVero, l.H1u2mIasnsuaect0iv7it,yJruelcyo-g2-023 nition from smart watch sensor data using a hybrid of principal component analysis and random forest algorithm November 2018 https://journals.sagepub.com/doi/10.1177/0020294018813692

[6] Chuanting Zhang, Dongfeng Yuan, Fast Fine-Grained Air Quality Index Level Prediction Using Random Forest Algorithm on Cluster Computing of Spark https://www.researchgate.net/publication/281061339

[7] Sajib Kabiraj, M. Raihan, Nasif Alv, Marina Afrin, Laboni Akter, Shawmi Akhter Sohagi, Etu Podder, Breast Cancer Risk Prediction using XGBoost and Random Forest Algorithm 2020 International Conference, https://ieeexplore.ieee.org/document/9225451

[8] Suneeta, V.B., Purushottam, P., Prashantkumar, K., Sachin, S., Supreet,

M. (2020). Facial Expression Recognition Using Supervised Learning. In: Smys, S., Tavares, J., Balas, V., Iliyasu, A. (eds) Computational Vision and Bio-Inspired Computing. ICCVBIC 2019. Advances in Intelligent Systems and Computing, vol 1108. Springer.

[9] Dai Chunni, SVM Visual Classification Based on Weighted Fea- ture of Genetic Algorithm 2015 sixth International Conference, https://ieeexplore.ieee.org/document/7462735

[10] Maniyar, H.M., Budihal, S.V. (2020). Plant Disease Detection: An Augmented Approach Using CNN and Generative Adversarial Network (GAN). In: Badica, C., Liatsis, P., Kharb, L., Chahal, D. (eds) Informa- tion, Communication and Computing Technology. ICICCT 2020. Com- munications in Computer and Information Science, vol 1170. Springer, Singapore.

[11] Pavaskar, S., Budihal, S. (2019). Real-Time Vehicle-Type Categorization and Character Extraction from the License Plates. In: Mallick, P., Balas, V., Bhoi, A., Zobaa, A. (eds) Cognitive Informatics and Soft Computing. Advances in Intelligent Systems and Computing, vol 768. Springer, Singapore.

[12] Suneeta V. Budihal, Rajeshwari M. Banakar, Evidence-based dynamic radio resource allocation to mitigate inter cell interference employing cooperative communication, IET Communications, Volume 14, Issue 12, July 2020, Pages 1848-1857