The open world assumption in model development indicates that a model may not have enough information to effectively manage data that is completely different or out of distribution (OOD). When a model encounters OOD data, its performance can deteriorate significantly. To enhance a model’s ability to handle OOD data, generalisation techniques can be employed, such as adding noise, which is easily implemented using deep learning methods. However, many advanced machine learning models are resource-intensive and optimised for use with specialised hardware like GPU, which may not always be accessible to users with limited hardware capabilities. To provide a deeper understanding and practical solutions for handling OOD data, this study explores detection, evaluation, and prediction tasks within the context of OOD on tabular datasets, specifically using common consumer hardware (CPUs). The study demonstrates how users can identify OOD data from available datasets and offers guidance on evaluating the selection of OOD data through straightforward experiments and visual-isations. Additionally, this study introduces a technique called Tabular Contrast Learning (TCL), a representation learning technique which is specifically designed for tabular prediction tasks. TCL achieves better results compared to heavier models while being more efficient, even when trained on non-specialised hardware. This makes it particularly beneficial for general machine learning users who face computational constraints. The results show that TCL outperforms other contrastive learning methods and various deep learning models in classification tasks.