Human gait has proved to be one of the standard biometrics for human identification. It is a non-invasive biometric method that uses human walking patterns specific for each human being. In most of the traditional methods, we use handcrafted features of simple convolutional models for gait analysis in human identification. Here we may face challenges addressing complex temporal dependencies in gait sequences. This study proposes a novel deep learning framework that applies multi-feature input representations. It combines Gait Energy Images (GEI), Frame Difference Gait Images (FDGI), and Histogram of Oriented Gradients (HOG) features. This is proposed for enhancing the accuracy of human identification. The proposed work implements a CNN-based feature extractor with an attention mechanism for gait recognition. The model is trained and validated on a labeled dataset, showcasing its ability to learn discriminative gait representations with improved accuracy. The proposed pipeline of activities include preprocessing and converting gait sequences into frames, organizing them using folder-based numerical extraction, followed by the training of an attention-enhanced convolutional network. The proposed model was found to perform better than existing methods on public datasets and works well even with different camera angles and clothing styles.