A logistic classification algorithm for the detection of heart failure.
SckitLearn, Pandas, Matplotlib
Dataset was an imbalanced one such that the number of positive cases is way less than the negative cases to the ratio 1:2. However, considering the cases of heart failure in a random population is hardly ever in the majority, the datasets weren't balanced as this will mess with real-life performance of the algorithm.
Because it's a health-related algorithm, it's sensible to want to improve recall to somewhere around upwards of 90%. Possible methods might be to:
- change the threshold of the decision function of the classifier
- changing the algorithm all together
- adding new features
- upsampling the positive cases