Analysis of Imbalance Learning Technique on Classification and Model Interpretability with LIME

Alzheimer's disease, a progressive neuro-degenerative disorder, poses a substantial global health challenge, necessitating accurate and timely diagnosis. However, the inherent class imbalance in Alzheimer's datasets presents a challenge for classification models leading to widespread delays in accurate and timely diagnosis. This research evaluates a comprehensive imbalance learning approach, the Synthetic Minority Over-sampling Technique (SMOTE) using eight different machine learning algorithms. The performance of models using SMOTE is compared and evaluated on a range of performance metrics like accuracy, recall, etc. Unexpectedly, results indicate that leveraging SMOTE decreases performance for most models, except for k-NN, where performance metrics are enhanced, although only marginally. The paper also uses explainable AI to demonstrate the predictions of the model interpretability. Notably, XG-Boost being the best performing model, achieves an accuracy of 86.66% (without SMOTE), underscoring its effectiveness in accurately diagnosing Alzheimer's disease, while its performance was negatively affected by SMOTE. This research contributes valuable insights for the development of classification tools in enhancing Alzheimer's disease diagnosis.