Gene Expression Pattern Classification for Cancer Subtype Prediction Using Microarray Data and Ensemble Learning

Microarray-based gene expression profiling identifies subtype-specific transcriptional signatures and supports precision oncology by representing cancer heterogeneity in high-dimensional molecular space. Dimensionality-induced overfitting, redundant gene representations, class imbalance, and poor generalisation across cancer subtypes plague conventional classification approaches. To overcome these limits, this research provides an Adaptive Gene Expression Ensemble Classification (AGEEC) framework for robust microarray cancer subtype prediction. Variance-aware gene filtering, mutual-information-driven feature prioritisation, and adaptive ensemble learning extract discriminative molecular fingerprints from high-dimensional expression profiles. To enhance subtype recognition, an optimised stacking architecture employs random forest, support vector machine, and gradient-boosting classifiers with a meta-classifier merging their probabilistic outputs. Layered hyperparameter optimisation and stratified cross-validation reduce selection bias and improve generalisation. Subtype Separation Efficiency (SSE) measures inter-subtype discrimination, Gene Signature Retention Rate (GSRR) biomarker reproducibility, and Perturbation Resilience Coefficient (PRC) prediction consistency under controlled expressionprofile perturbations. Benchmark cancer microarray datasets are used to evaluate AGEEC against individual classifiers and ensemble baselines. The framework aims for an SSE of 97.8%, GSRR of 94.6%, PRC of 96.3%, and a 15.2% improvement in subtype discrimination over the baseline.