SAGB: A Hybrid Stacked AutoEncoder-Gradient Boosting Model for Accurate Classification of Amyotrophic Lateral Sclerosis (ALS)

Amyotrophic Lateral Sclerosis (ALS) is a progressive neurodegenerative disorder that affects motor neurons, leading to muscle weakness and, eventually, respiratory failure. Early diagnosis is critical for managing ALS, but it remains a challenging task due to the disease's complex presentation and the variety of factors involved, including genetic, clinical, and demographic variables. In this study, we propose a hybrid machine learning approach for accurate classification of Amyotrophic Lateral Sclerosis (ALS) patients using a Stacked Autoencoder for feature extraction and a Gradient Boosting Model (GBM) for classification. The ALS dataset, sourced from Kaggle, contains clinical, genetic, and demographic features, which we preprocess by normalizing and reducing dimensionality using the AutoEncoders bottleneck layer. The extracted features fed into the GBM for classification. We optimize the GBM through Grid Search to fine-tune hyperparameters such as learning rate, number of estimators, and tree depth. Our results show that the Autoencoder + GBM model outperforms traditional classifiers like Logistic Regression (LR), Support Vector Machines (SVM), k-Nearest Neighbors (kNN), and Random Forests (RF) across several evaluation metrics. Specifically, the Autoencoder + GBM achieves an accuracy of 93.8%, with a precision of 91.2%, recall of 92.3%, and an Fl-score of 91.7%. These results indicate the effectiveness of feature extraction via AutoEncoders in improving the performance of complex classification tasks like ALS diagnosis. Comparatively, simpler models such as LR and Naive Bayes (NB) show significantly lower accuracy, around 85.4% and 83.3%, respectively.