Feature selection-based prediction model using binary dung beetle optimisation algorithm and improved CNN with regularisation model

Feature selection is a crucial process in data analysis, aiming to identify and eliminate irrelevant features to enhance model performance and efficiency. This research proposes a novel approach to address limitations in existing methodologies, such as overlooking feature interactions and struggling with large-scale datasets. Firstly, comprehensive data preprocessing steps including cleaning, transformation, and organisation are implemented to enhance data quality. Next, a binary version of the dung beetle optimisation algorithm is introduced to tackle feature selection issues, leveraging a U-shaped transfer function. Additionally, the dung beetle optimisation algorithm is augmented with Lévy flight to enhance solution diversity and a local search algorithm to prevent entrapment in local optima. The convolutional neural network is enhanced for prediction using dynamic chunk-based max pooling and dung beetle optimisation algorithms for layer improvement and weight refinement. Finally, to mitigate overfitting and optimise overall performance, a hybrid L1/2 + L2 regularisation model is applied.