<p>Fault identification in injection systems is essential for addressing issues such as low power, intermittent combustion, misfiring, and excessive fuel consumption. This study employed feature selection techniques and ensemble learning to enhance the detection of injection failures in electronic diesel engines by diagnosing faults in the left and right injection benches. The research addresses a gap in fault detection and diagnosis for offshore diesel engines by applying ensemble learning with a heterogeneous approach, combining three feature selection methods to improve the performance of four classifiers. The results obtained highlight the relevance of the proposed methodology. The primary objective was to develop machine learning models for fault classification in the fuel injection system of the electronic diesel engine model 3512B, commonly used in offshore platform applications. Four classification algorithms were investigated: KNN, SVM, random forest, and RUSBoosted tree. All 40 features available in the engine monitoring system were initially tested with each algorithm. Subsequently, the most relevant features were extracted using minimum redundancy maximum relevance, Chi-squared, ANOVA, and a committee approach that combined all three methods. In total, 49 models were evaluated. The best performance was achieved by an SVM model using 10 features, with an F1-Score of 97.89%. The second-best model was a random forest, achieving an F1-Score of 97.72% with only two features selected through the committee method. These findings demonstrate the feasibility of accurately identifying injection faults in this marine diesel engine using only exhaust temperature data from the left and right sides.</p>