Stroke prediction using deep learning and transfer learning approaches

Stroke prediction using deep learning and transfer learning approaches

ABSTRACT

Stroke is a leading cause of death and disability worldwide, and early detection is crucial for effective treatment and improved patient outcomes. This project focuses on developing a predictive model for stroke risk using machine learning and deep learning techniques. The dataset used contains various patient features such as age, gender, hypertension, heart disease, smoking status, and clinical measurements like average glucose levels and BMI. In the proposed system, multiple algorithms, including XGBoost, Random Forest, Decision Trees, AdaBoost, and several deep learning models such as Feedforward Neural Networks (FNN), Long Short-Term Memory (LSTM), and TabNet, are evaluated to predict stroke risk with high accuracy. The models were trained on a preprocessed dataset, where missing values were handled, and categorical variables were encoded using label encoding. Additionally, resampling techniques like SVMSMOTE and RandomUnderSampler were used to address class imbalance. The results of the proposed system showed that Random Forest and FNN achieved the highest performance, with accuracies of 98.72% and 97.90%, respectively. This system offers a reliable tool for early stroke detection and can assist healthcare professionals in making informed decisions for timely interventions.