Prediction of train delays is critical for better adjustment in the plan made at stations on one hand and also played a huge role in gaining customer satisfaction. In this paper, a more detailed methodology approach has been suggested for train delay prediction using SVM with MZ-Score preprocessing and LR feature selection. The MZ-Score is used to flag and prune outliers in the data, with LR carrying out feature selection by reducing coefficients towards zero for less relevant features. The suggested way is implemented and tested on the Train_delay_Prediction. Csv dataset. This evaluation tests how much the performance of different FS methods—RFE, PCA and LR with MZ-Score preprocessing can change. The results demonstrate that better performance and balance between precision, recall can be achieved by the synergy of MZ-Score preprocessing, LR feature selection and SVM for train delay prediction. From the results obtained its proved that proposed MZ-Score + LR + SVM gave accuracy of 96.45%, precision of 0.94 and recall of 0.92 which is high compared to other existing algorithms.