The main goal of this thesis is to evaluate individualized treatment effects (ITE) using neural networks. The individuals are university students and treatment is defined as attendance in Supplemental Instruction (SI). The variable of interest is given by the grade at the end of the semester. So the main question of this thesis is, how much each individual student would improve by attending SI. Even though past data is available, each student can either attend SI, or not. Hence only the grade at the end of the semester for one of the two scenarios is known, such that the true ITE will always be unknown. Therefore statistical methods need to be implemented in order to estimate the ITE. While we consider several methods, the main focus lies on neural networks. A simulation study is used to evaluate the accuracy of the statistical methods developed. The methods are applied on artificially generated data, where the true relationship between the variables is known and the true ITE can be calculated. The simulation study showed that neural networks, along with random forest, performed best in estimating the ITE. Since individual neural networks showed quite a bit of variation, we used an averaging method to derive the final prediction model.The application data for this analysis contains several thousand observations, each describing one student. For each student, more than 20 covariates are available with both university related information, such as major or term GPA and personal information, such as gender and age. The main finding in this analysis is that students with weaker academic background benefit more by services like SI. Furthermore, we may use pre-semester data such as high school GPA, campus GPA, and academic status to create an early warning system to encourage students into the SI program.