The problem of employee attrition in every organisation is concerning the employee turnover ratio thereby increasing the cost of investment in human resources. Various factors are reasonable for the rapid attritions at different phases. The purpose of the current study is to predict the employees who are likely to leave the organisation. The factors that lead to attrition are identified using random forest algorithm. Random forest algorithm is a widely used supervised machine learning technique for classification and prediction. However, the random forest algorithm has certain issues like it is too slow and ineffective for real-time predictions, i.e., the large number of trees can make the algorithm, which results in slower model. Therefore, the study proposes, a new alternative for choosing the appropriate decision trees based on the concept of fractional factorial design of experiments. The different performance criteria were compared across the modified random forest algorithm, existing random forest algorithm, support vector machine (SVM), logistic regression (LR), Naïve Bayes (NB), K-nearest neighbour (K-NN), decision tree, XGBoost tree and neural network (NN). It was found that the modified random forest algorithm excelled in all criteria and performed better than the existing ones.