A predictive analytics approach for stroke prediction using machine learning and neural networks

Dev, Soumyabrata; Wang, Hewei; Nwosu, Chidozie Shamrock; Jain, Neeraj; Veeravalli, Bharadwaj; John, Deepu

doi:10.48550/arxiv.2203.00497

Cited by 1 publication

(3 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The most important features for stroke prediction are age, marital status, hypertension, gender, work type, and average glucose level. (Dev et al 2022).…”

Section: Datasetmentioning

confidence: 99%

“…Recognizing and removing redundant features that can be safely ignored without sacrificing prediction model performance would both assist clinicians in diagnosing stroke and reduce the computational cost of training. However, prior research on the impact of risk factors on stroke prediction using Principal Component Analysis (Nwosu et al 2019), Learning Vector Quantization (Dev et al 2022), and experimentation with different feature combinations (Dev et al 2022) show that the features are not highly correlated. Therefore, all patient attributes should be used to train our stroke prediction model.…”

Section: Datasetmentioning

confidence: 99%

“…In this paper, we analyze the performance of TabTransformer in predicting the occurrence of brain stroke. We then benchmark other state-of-the-art methods: the Kaggle competition winner for the stroke prediction dataset 1 , XGBoost (Chen and Guestrin 2016), and the model with the most performance in a 2019 (Nwosu et al 2019) and 2022 (Dev et al 2022) study, Multilayer Perceptron.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Analyzing the Performance of TabTransformer in Brain Stroke Prediction

Xia¹,

Ramezani

2023

J Stud Res

View full text Add to dashboard Cite

The adoption of electronic patient health records has paved the way for machine learning anddeep learning in disease diagnostics and prediction. Though traditionally tree-based algorithmshave performed well on structural data, neural networks are known to perform well onunstructured data and data with a large number of input features. Furthermore, transformer-based models such as TabTransformer have been shown to perform competitively with tree-basedalgorithms (Huang et al. 2020). In this paper, we compare TabTransformer’s performance withother state-of-art machine learning algorithms such as XGBoost, RandomForest, DecisionTree, andfeed-forward Multilayer Perceptron. We discovered that TabTransformer shows no significantimprovement over MLP and performs worse in certain metrics. Neither TabTransformer nor MLPperformed better than XGBoost, the best-performing algorithm for brain stroke prediction inKaggle competitions.

show abstract

“…The most important features for stroke prediction are age, marital status, hypertension, gender, work type, and average glucose level. (Dev et al 2022).…”

Section: Datasetmentioning

confidence: 99%