Increasingly complex modern diesel engines have made modelling tools essential for development and control. The computational cost of physics-based models limits their use in real-time applications, motivating the adoption of data-driven approaches. This thesis presents the development of predictive models for estimating key engine output variables using experimental data collected from a test bench session. Linear regression models were first developed and analysed. Subsequently, nonlinear approaches, including quadratic and power law models, were introduced to capture more complex relations. Feedforward neural networks were then explored, including multi-output architectures. The proposed models were evaluated and compared in terms of predictive accuracy.