The aim of this work is to predict high education students’ progress and achievement by forecasting their final grades in any taught courses as early as possible during a study semester or term, using an innovative neural network shrinking technique. The trained neural network NN is divided into a head and several tails in a cascaded sequential manner. Course data from previous offerings are used in a network which is trained using a feedforward BP scheme to extract input and output weight coefficients and biases. One of the input features is reduced by a fraction of its original value and used with the same input data in a tailed NN, which is initiated with the extracted coefficients and biases from the previous run. The training is continued in a cascaded manner until eliminating one input assignment. The whole process is continued for other assignments to be eliminated. This algorithm can be constituted as a dynamic process workbench for an alternative method of forecasting the achieved grades of high educational students in an easy and cost-free manner. The earlier it is to forecast final grade the easier it is to alleviate outcomes of course grading. The procedure is applied on two different courses offered by the same teacher. The input data of different batches of students attending a particular course are used. It is found that a tentative accuracy of predicting final grades from the start is possible.