Shot‐Sense Cricket: An Interpretable Hybrid Framework Combining Machine Learning and Efficient‐Net for Robust Cricket Shot Classification

Identifying cricket shots from static pictures represents a difficult assignment because similarities between classes are high, the body position of the player is unclear, stadium surroundings are messy, and the perspective of the camera changes frequently. Recognition systems often give the wrong names to movements like drives and leg‐glances, which have nearly the same body postures and bat positions. Stadium side camera angles, blockage from safety gear, and busy stadium backgrounds make this difficulty worse. The study introduces shot‐sense cricket, which is a clear hybrid framework for cricket shot classification. Instead of looking at Efficient‐Net, XG‐Boost, SHAP, Media‐Pipe, SIFT, ORB, and GLCM as separate instruments, the framework brings a method that was checked with care. This method uses a confidence‐weighted late‐fusion architecture that joins pose‐based details, surface patterns, and local‐invariant features with deep Efficient‐Net embeddings. This architecture is joined with a SHAP‐driven diagnostic layer. This layer follows single wrong classifications back to the parts of the image that caused the errors. It has been observed that specific image regions are responsible for these mistakes. Totaling 4724 static images, the collection covers four cricket shot classes (drive, pull, sweep, and leg‐glance/flick), Many tests were done on four groups of hand‐made features, six classification programs, and three Efficient‐Net versions. These tests show that the adjusted Media‐Pipe + XG‐Boost setup reaches a 98.0% test accuracy. The proposed late‐fusion framework reaches a 97.5% accuracy while showing better strength against background noise. The proposed framework offers a scalable, computationally efficient alternative to purely deep learning pipelines for image‐based cricket analytics and lays the groundwork for future video‐based and temporal (e.g., LSTM‐based) extensions.