**3. Literature survey**

The focus of the research work presented in this paper is on the detection of faults due to attacks and the methods used to detect and classify the data.

Zainib Noshad, Nadeem Javaid, Tanzila Saba [1] The use of wireless sensor networks (WSNs) in a variety of environments makes them susceptible to errors. Unstable and dangerous conditions. This makes WSN vulnerable to errors in software, faults in hardware and communication. Fault detection in WSNs has become a challenging task because of the sensor's constrained resources and varied deployment environments. The classification of gain, offset, spike, data loss, out of bounds, and stuck-at faults at the sensor level is done using Support Vector Machine (SVM), Convolutional Neural Network (CNN), Stochastic Gradient Descent (SGD), Multilayer Perceptron (MLP), Random Forest (RF), and Probabilistic Neural Network (PNN) classifiers. Two of the six faults—the spike and data loss faults—are brought on by the datasets. The Detection Accuracy (DA), True Positive Rate (TPR), Matthews Correlation Coefficients (MCC), and F1-score are used to compare the results. Simulations demonstrate that the RF method achieves a higher rate of defect detection than the other classifiers.

Salah Zidi, Tarek Moulahi, and Bechir Alaya [2] one of the easiest ways to find failure in WSNs appears to be to use machine learning. SVM is employed in our context to define a decision function, which is based on statistical learning theory. This technique has a lot of potential for multidimensional data learning in addition to having demonstrated performance in a number of fields. This method, which makes use of kernel functions, has a significant capacity for adaptability for nonlinear classification scenarios, such as our case of fault detection. This has the potential to be very helpful in fault prevention. The goals of this research are to use a dynamic classification approach to track sensor activity through its data in order to predict errors as quickly as feasible in the same context of prevention.

Terry Windeatt [3] Multilayer Perceptron (MLP) classifier settings can be difficult to adjust, as is widely known. In this study, a metric that can forecast how many classifier training iterations will take to get the best results from an ensemble of MLP classifiers is described. The measure, which is based on a spectral representation of a Boolean function, is computed between pairs of patterns on the training data. With this representation, accuracy and diversity can be combined into a single statistic that describes the mapping from classifier decisions to the target label.

Luofan Dong, Huaqiang Du, Fangjie Mao [4] Convolutional neural networks (CNNs) recently demonstrated outstanding performance in a variety of applications, including computer vision and remote sensing semantic segmentation. Much interest is focused on the capacity to learn CNN's high-representational properties. On the other hand, the random forest (RF) technique is frequently used for variable selection, classification, and regression. This article tested a technique based on the fusion of an RF classifier and the CNN for a very high-resolution remote sensing (VHRRS) based forests mapping. This method was based on the previous fusion models that fused CNN with the other models, such as conditional random fields (CRFs), support vector machine (SVM), and RF. Huwaida T. Elshoush, Esraa A. Dinar [5] Spam prevalence is rising daily as electronic emails are used more frequently. As a result, spam emails have grown to be a serious issue that reduces the use of electronic emails for communication. Several machines learning approaches, including Naive Bayes (NB), K-Nearest Neighbor (KNN), Support Vector Machine (SVM), Artificial Neural Network (ANN), and Decision Tree, provide email spam filtering solutions (DT). This study examines various machine learning methods, namely Adaboost and Stochastic Gradient Descent, to filter spam emails (SGD).

Mrutyunjaya Panda, Ajith Abraham [6] Security of network traffic is growing to be a significant issue for computer networks as the internet expands. The frequency of attacks on the network is rising over time. Such network attacks are nothing more than intrusions. The network and data have been protected against threats by using intrusion detection systems to identify intrusions. Large amounts of network data are monitored, analyzed, and classified into abnormal and regular data using data mining algorithms. Poornima G, K Suresh Babu, K B Raja, K R Venugopal, and L M Patnaik [7] proposed to find the probability of correctly identifying a faulty node for three different types of faults based on normal bias. The nodes fault status is declared based on its confidence score that depends on the threshold valve. Uma R. Salunkhe, Suresh N. Mali [8] used an intrusion detection system (IDS) to detect hostile activity has been an efficient technique to increase security. An intrusion detection system is anomaly detection. Due to its inability to accurately detect all sorts of attacks, current anomaly detection is frequently characterized by high false alarm rates and only modest accuracy and detection rates. Using the KDD-99 Cup and NSL-KDD datasets, a test is run to assess how well the various machine learning methods perform.

Miao X, Liu Y, Zhao H, Li C [9] system which detects the attacks in the wireless sensor network the KDD cup 99 data set is used in the present paper and the to classify the attacks in the WSN's the KNN classifier is used, But the detection rate achieved with this classifier is very poor and the highest detection rate is 75% and that is for k = 5. Gharghan S.K, Nordin R, Ismail M, Ali J. A [10] a hardware model for intrusion detection system is suggested this model has failed to give the accurate result, due to some hardware vulnerabilities and it is complex to design and human intervention is required.

In [11] the authors have discussed Intrusion detection system and used Decision tree, SVM, MLP algorithm. The result shows that MLP outperforms the other classifier with accuracy of 91%. In [12] the authors elaborate on layer wise DoS attack and its defense mechanisms and classification. In [13] the authors detect faults in WSN using hidden Markov model, KDD cup 99 data set is used, the accuracy they have achieved for test data is 77.11%. In paper [14] fault detection using deep learning algorithms is done. KDD cup 99 data set is used and MLP, SVM algorithms are used and the accuracy is 91%. In [15] the fault detection in WSN using Internet of things based on improved BP Neural network Leven berg- Marquard algorithm is applied with a accuracy result of 91%.

From the papers surveyed, for selecting feature subset Recursive feature elimination method is implemented. All the independent variables in supervised learning is known as features of the data. Elimination in this context means eliminating the features. Doing a process repetitively to eliminate the features of the data is known as Recursive feature elimination. KDD is a type of data set and an online repository that contains data from all different sorts of intrusion attempts. It mainly includes DOS, R2L, U2R, and PROBE. RF, SVM, SGD, MLP classifiers will be assessed on the KDD dataset in this research.

*Efficient Machine Learning Classifier for Fault Detection in Wireless Sensor Networks DOI: http://dx.doi.org/10.5772/intechopen.111462*
