This paper evaluates the use of Machine Learning (ML) Binary Classification as an innovative approach for assessing subsea pipeline freespan integrity, offering an alternative to the traditional DNV RP F105 code and standards. The study investigates the performance of various ML algorithms: Logistic Regression, Decision Tree, Random Forest, and Neural Network, applied to the binary classification of pipeline freespan acceptance.

The methodology begins with data collection, leveraging datasets generated from pipeline freespan assessments based on calculated conventional method. The data undergoes pre-processing to ensure consistency and quality, involving steps such as data cleaning, normalization, and feature engineering.

The pre-processed data is partitioned into training and testing subsets. The training set is used to train the models, where each algorithm learns to identify patterns and correlations in the data. Then hyperparameter tuning is performed during this phase to optimize model performance.

Subsequently, the validation set is employed to fine-tune the models, refining parameters to improve accuracy. Finally, the test set is used for comprehensive evaluation, where metrics such as confusion matrix, precision, and True Negative Rate are calculated to assess model performance and reliability.

The results are analysed and compared to identify the most suitable algorithm for pipeline freespan assessment. The findings highlight the potential of ML-based approaches to enhance the accuracy and efficiency of subsea pipeline integrity evaluations, presenting a viable alternative to conventional methods.