<p>Internal combustion (IC) engine generates power by burning fuel inside a combustion chamber. However the complex patterns in engine data can be challenging, as machine learning models may struggle with data sparsity. Deep learning models may struggle with real-time fault detection due to their performance can be sensitive to domain shifts, requiring frequent recalibration for varying engine conditions. To overcome these challenges, introduce the Modified Fast Gated Recurrent Neural Network (MFGRNN) for IC engine fault detection. Sensor data collection is the first step in the IC engine fault detection process. It is preprocessed using Gaussian Random Incremental Principal Component Analysis (GRIPCA) to remove redundancy, Transformer-Enabled Generative Adversarial Imputation Network (TE-GAIN) to impute missing data, and Sigmoid Normalization Method (SNM) to standardize the data. Modified Sparse and Low Redundant Subspace Learning based Dual Graph Regularized (MSLSDR) with a modified Least Absolute Shrinkage and Selection Operator (LASSO) regularization is used for feature selection in order to find the pertinent features. These features are input to the Modified Fast Gated Recurrent Neural Network (MFGRNN) for fault detection. Switching from Batch Normalization to Layer Normalization improves hidden state stability and enhances prediction accuracy in recurrent networks. The system classifies engine conditions as normal or faulty. If a fault is detected, it is further classified as either critical or minor for timely maintenance. The method achieves a Precision of 94.65%, a Specificity of 97.29%, and an Error of 3.6% for fault detection, demonstrating strong performance across these metrics. The proposed approach will benefit from more precise, real-time diagnostics, minimizing maintenance costs and downtime.</p>