Scholarlink

<p>Alzheimer’s disease (AD) is a progressive neurodegenerative disorder where early diagnosis is crucial. Conventional methods like PET scans and CSF analysis are accurate but invasive, costly, and impractical for large-scale screening. To address these limitations, this study proposes a non-invasive, machine learning-based approach for early AD detection using handwriting biometrics. The methodology utilizes the DARWIN dataset, comprising handwriting samples from 174 subjects performing 25 tasks designed to evaluate cognitive and motor functions. Multiple classification algorithms, including Support Vector Machines (SVM), Logistic Regression (LR), Random Forest (RF), and XGBoost, were evaluated. Ensemble-based classifiers such as RF and XGBoost outperformed other models, achieving classification accuracies exceeding 90%. To address the moderate class imbalance in the dataset, where the numbers of AD and healthy control subjects are not equal, we adopted a hybrid strategy that combines both data-level and algorithm-level techniques. Specifically, the Synthetic Minority Over-sampling Technique (SMOTE) was applied to generate synthetic examples of the minority class, thus balancing the dataset before model training. Simultaneously, Recursive Feature Elimination (RFE) minimizes the impact of noise or redundant features while reducing feature dimensionality and enhancing class separability. Next, a stacked ensemble model is created, with a Random Forest serving as the meta-classifier and LR, XGBoost, LightGBM, and CatBoost serving as base learners. This proposed model demonstrated resilience and good predictive performance with an AUC of 0.99 and an accuracy of 99.3%. According to these results, handwriting-based machine learning presents a non-invasive, scalable, and economical method for early AD identification, with substantial clinical integration potential.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Novel Non-invasive Handwriting-Based Machine Learning Approach for Early Detection of Alzheimer’s Disease

  • 期刊论文

  • 发表日期：

    2025年10月14日

查看全文
  • Rajib Saha,
  • Anirban Mukherjee,
  • Abhishek Bal,
  • Adrita Chakrabarti,
  • Shrayanendra Nath Mandal

摘要

Alzheimer’s disease (AD) is a progressive neurodegenerative disorder where early diagnosis is crucial. Conventional methods like PET scans and CSF analysis are accurate but invasive, costly, and impractical for large-scale screening. To address these limitations, this study proposes a non-invasive, machine learning-based approach for early AD detection using handwriting biometrics. The methodology utilizes the DARWIN dataset, comprising handwriting samples from 174 subjects performing 25 tasks designed to evaluate cognitive and motor functions. Multiple classification algorithms, including Support Vector Machines (SVM), Logistic Regression (LR), Random Forest (RF), and XGBoost, were evaluated. Ensemble-based classifiers such as RF and XGBoost outperformed other models, achieving classification accuracies exceeding 90%. To address the moderate class imbalance in the dataset, where the numbers of AD and healthy control subjects are not equal, we adopted a hybrid strategy that combines both data-level and algorithm-level techniques. Specifically, the Synthetic Minority Over-sampling Technique (SMOTE) was applied to generate synthetic examples of the minority class, thus balancing the dataset before model training. Simultaneously, Recursive Feature Elimination (RFE) minimizes the impact of noise or redundant features while reducing feature dimensionality and enhancing class separability. Next, a stacked ensemble model is created, with a Random Forest serving as the meta-classifier and LR, XGBoost, LightGBM, and CatBoost serving as base learners. This proposed model demonstrated resilience and good predictive performance with an AUC of 0.99 and an accuracy of 99.3%. According to these results, handwriting-based machine learning presents a non-invasive, scalable, and economical method for early AD identification, with substantial clinical integration potential.