Fault-Tolerant Algorithm for Software Preduction Using Machine Learning Techniques

Introduction

SRGMs contain two basic type of input data are TDD and IDD. TDD describes the recoding of failure data in separate time and the IDD tells the cumulative number of failures occur over a given period of time. For solving the problem with good approach, we need to work on Software testing in early phase for the reliable software. Different types of data need to be selected to predict the performance of model and one of them is SVM approach which is based on GA (JH Lo, 2010). Other approach is RVM based approach for predicting reliability of software having two features; first model uses only recent failure data out of available data. Secondly uses cumulative failure data instead of inter-failure data. RVM learning scheme is applied to the failure time data, it recognizes the inherent internal temporal property of the software failure in order to capture the more current hidden feature of software which is applied to failure time data and forcing the network to learn. RVM approach achieves the best performance in terms of accuracy. Cumulative failure data normally have less sharp change points than inter-failure data having stable pattern and capture easily (Jun-gang L et al., 2009). Some well-know ML techniques are used for predicting the reliability of software such as ANN, SVM, CNN, DT’s, FIS. The importance of each method and the capability of reliable models for comparative analysis have been done. For varied real life sets of failure data, precision is very effective and for output data as day of failures. K-fold is used for input-output datasets for training and validation of the model (P Kumar & Y Singh, 2012). Other real-life data used for the S-shaped growth model, the inflection S-shaped and hyper-exponential model for predicting the reliability. Exponential growth model fits are observed data very well. While In case of calendar time-based data non-homogeneous time distribution of the testing effort, exponential growth model doesn’t fit the observed data. The hyper-exponential growth model gives a better estimate for the number of faults for ordinary exponential growth model. In both cases the curve is similar, but in case of hyper-exponential model estimation results is unrealistic and inaccurate (Mitsuru Ohba, 1984). Software reliability model are stochastic and dynamic; The Software always need improvement. So complete testing is not possible, sufficient testing and the proper maintenance improves the reliability of software. In this situation proper CASE study is performed to build the product such as requirement analysis, modelling design, program construction and the testing. The main focus is to how to improve the quality of software based on random phenomenon and the activities to get the desired output or reliable outcome (A Quyoum et.al, 2010).