Interpretable machine learning addresses the challenge of understanding how complex models make decisions. Despite many approaches, a comprehensive understanding of achievements and challenges remains lacking. This survey summarizes existing techniques to enhance model interpretability, discusses key issues for future research, such as user-friendly explanations and evaluation metrics. Interpretable machine learning is divided into intrinsic and post-hoc categories. Intrinsic models are designed to be inherently interpretable, such as decision trees and linear models. Post-hoc methods use additional models to explain existing ones. Global interpretability provides an overview of model behavior, while local interpretability focuses on individual predictions. Intrinsic models include globally interpretable models, such as decision trees, and locally interpretable models, such as those using attention mechanisms. Post-hoc global explanations aim to understand model behavior, while post-hoc local explanations focus on individual predictions. Techniques like feature importance and activation maximization are used for explanations. Traditional machine learning explanations often rely on feature engineering, while deep learning explanations focus on understanding neural network representations. Methods like permutation importance and attention mechanisms are used to explain model decisions. Local explanations, such as attribution methods, identify feature contributions to predictions. Techniques like back-propagation, mask perturbation, and investigation of deep representations are used to generate explanations. Applications of interpretable machine learning include model validation, debugging, and knowledge discovery. Challenges include designing effective explanation methods and evaluating their faithfulness. Future directions involve creating more user-friendly explanations that are understandable and helpful for end-users.Interpretable machine learning addresses the challenge of understanding how complex models make decisions. Despite many approaches, a comprehensive understanding of achievements and challenges remains lacking. This survey summarizes existing techniques to enhance model interpretability, discusses key issues for future research, such as user-friendly explanations and evaluation metrics. Interpretable machine learning is divided into intrinsic and post-hoc categories. Intrinsic models are designed to be inherently interpretable, such as decision trees and linear models. Post-hoc methods use additional models to explain existing ones. Global interpretability provides an overview of model behavior, while local interpretability focuses on individual predictions. Intrinsic models include globally interpretable models, such as decision trees, and locally interpretable models, such as those using attention mechanisms. Post-hoc global explanations aim to understand model behavior, while post-hoc local explanations focus on individual predictions. Techniques like feature importance and activation maximization are used for explanations. Traditional machine learning explanations often rely on feature engineering, while deep learning explanations focus on understanding neural network representations. Methods like permutation importance and attention mechanisms are used to explain model decisions. Local explanations, such as attribution methods, identify feature contributions to predictions. Techniques like back-propagation, mask perturbation, and investigation of deep representations are used to generate explanations. Applications of interpretable machine learning include model validation, debugging, and knowledge discovery. Challenges include designing effective explanation methods and evaluating their faithfulness. Future directions involve creating more user-friendly explanations that are understandable and helpful for end-users.