Badminton is a popular and fastest racquet sport that is played in the world. The current methods that are available for analyzing badminton matches can be costly and are not suitable for the small-scale badminton matches. Hence, there are a lot of techniques or methods that are used to track and detect the badminton shuttlecock in badminton matches. As a result, the machine learning methods have the potential to automate and enhance the analysis the badminton shuttlecock tracking and line detection. The low-cost implementation of data collection, which involves recording badminton videos using a phone camera, provides an affordable option for analyzing small-scale badminton matches. This research proposed an affordable badminton shuttlecock tracking and line detection approach specifically designed for small-scale badminton matches. In this research, the performance of the TrackNet Model I and TrackNet Model II are compared in badminton shuttlecock tracking and line detection. The accuracy, precision and recall are metrics used to evaluate and compare performance in this study. This research uses a structured technique that includes the collecting of dataset, model architecture, identification of badminton court lines, and model implementation. According to the performance metric, TrackNet Model II achieves 100% recall accuracy, precision accuracy of 95.45%, and accuracy of 95.45%. With the custom dataset, the TrackNet Model II performs better when tracking the badminton shuttlecock.