Alzheimer's disease prediction. Team 1

Preview:

Citation preview

What is Alzheimer'sDisease ?

The Alzheimer'sDiseasePredictionOfLongitudinalEvolution

What is TADPOLE Challenge?

Goal:Predict how many days does it takes to convert to AD

Current State Days = ? AD

Pipeline

8,892 Observations

One observation is one visit

1,907 Features

RAW DATA

1,200 Observations

One observation is one visit

132 Features

CLEAN DATA

Target feature:

Time_To_Convertion

Feature selection

1. By expert in Alzheimer Disease area

2. Dimensionality Reduction Techniques

Explained variance of PCA components

Model selection and estimation

Select estimators ⇨ Fit and train ⇨ Compare ⇨ Select optimal

Support Vector RegressorLinear SVRLassoRidge RandomForestRegressorXGBoost

Model selection and estimation

Select estimators ⇨ Fit and train ⇨ Compare ⇨ Select optimal

Support Vector RegressorLinear SVRLassoRidge RandomForestRegressorXGBoost

Our choice

RandomForestRegressorMAE: 404.841703713 Best params: {'model__max_features': 'sqrt','model__max_depth': 8,'criterion' : 'mse'}

Results

Error of prediction in days

±404Based on one visit in scope of decades

experiencein data science project with real issues

contributiontowards AD data understanding

Results

Mean absolute error

Our small

Gained team

thanks!Jupyter Notebook link:

http://tinyurl.com/y8tsn4q9

Recommended