Alzheimer’s disease is a neurodegenerative disorder that significantly impacts cognitive functions. Early and accurate diagnosis is crucial for effective intervention and patient care. This study evaluates and compares the performance of various machine learning and deep learning models on the Kaggle dataset to identify the most accurate AI-based approach for AD detection and validate the effectiveness of each method. The study utilized a publicly available Kaggle dataset containing health information for ۲,۱۴۹ patients, uniquely identified by IDs ranging from ۴۷۵۱ to ۶۹۰۰. This dataset included demographic details, lifestyle factors, medical history, clinical measurements, cognitive assessments, and Alzheimer’s disease diagnosis. Initially, the dataset was normalized using Python to enhance model compatibility and then split into training and testing subsets, with ۸۰% allocated for training and ۲۰% for testing. Subsequently, various machine learning algorithms (KNN, LR, SVM, RF, DT) and deep learning models (CNN, LSTM) were applied to the normalized dataset. Ultimately, model performance was evaluated using accuracy, precision, and recall scores calculated on the test data. Among the machine learning algorithms, Decision Tree (DT) and Random Forest (RF) demonstrated strong performance, achieving accuracy, precision, and recall scores of (۹۵, ۹۶), (۹۳, ۹۶), and (۹۳, ۹۱), respectively. However, deep learning models (CNN, LSTM), achieving accuracy, precision, and recall scores of (۸۹, ۹۱), (۸۸, ۹۰), and (۸۸, ۹۰), respectively. The comparative study indicated that traditional machine learning algorithms, particularly Decision Tree (DT) and Random Forest (RF), outperformed deep learning models (CNN, LSTM) in the context of Alzheimer's disease classification. DT and RF achieved the highest accuracy, precision, and recall scores—up to ۹۶%—highlighting their effectiveness in structured tabular data scenarios. In contrast, while deep learning models demonstrated competitive performance with scores reaching ۹۱%, they slightly lagged behind classical models. This suggests that for this specific dataset, tree-based models may offer a more efficient and interpretable solution compared to complex deep architectures.
