ASD-Pipeline: An Ensemble Machine Learning Framework Integrating Feature Selection, Behavioural Clustering, and Class Rebalancing for Accurate Autism Spectrum Disorder Prediction

Approved

Relevance and Originality

The proposed work addresses a key issue in autism diagnostics: the fragmented nature of existing machine learning models for ASD detection. By consolidating multiple underutilized strategies—adaptive feature selection, behavioral clustering, and synthetic data augmentation—the study offers a cohesive and forward-thinking framework. This integrative approach stands out for its originality, especially in a field where such comprehensive models remain rare. The focus on real-world diagnostic limitations elevates its practical importance and scientific contribution.

Methodology

The methodology demonstrates strong conceptual depth and technical precision. Each of the five stages is well-justified and sequentially supports the pipeline’s goal of improving ASD prediction. The combination of feature selection layers in FlexiFeat is particularly notable, enhancing data quality before downstream modeling. Using hybrid clustering techniques to identify behavior patterns adds interpretability, while Cluster-SMOTE for addressing class imbalance demonstrates thoughtful problem framing. The final ensemble classifier setup ensures predictive consistency and model resilience, confirming a robust engineering design.

Validity & Reliability

The high accuracy, specificity, and F1-score suggest strong internal validity, supported by the use of a wide range of evaluation metrics, including MCC. The model’s structure inherently supports generalization by eliminating irrelevant data and emphasizing class balance. However, the reliability could be reinforced by reporting results from multiple datasets or through cross-institutional validations. The pipeline’s modular design increases confidence in its adaptability, but external testing would further confirm its clinical or cross-domain applicability.

Clarity and Structure

The research is structured with clear logic and technical coherence. Each phase of the pipeline is explained with appropriate terminology and flow, facilitating reader understanding even for interdisciplinary audiences. The narrative remains grounded in both the medical and machine learning context, which improves its accessibility. A slight improvement could be made by including illustrative elements such as diagrams or flowcharts to depict the end-to-end architecture of the ASD-Pipeline, especially to show interaction between feature selection, clustering, and classification stages.

Result Analysis

The outcome discussion is strong and effectively illustrates the superiority of the ASD-Pipeline over prior models. The reported improvements across all metrics are not only statistically impressive but also practically meaningful in clinical prediction tasks. The model’s consistent performance across accuracy, precision, and specificity reflects a balanced and reliable prediction framework. The interpretation of these results is aligned with the initial research objectives and affirms the strength of integrating behavioral insights with advanced machine learning methods.