Classification problem using SVM, decision tree, random forest and KNN, measuring the accuracy of each one (the best was random forest, with an accuracy of 98%). The goal is to predict if, given some medical conditions, a horse can survive or not. The dataset is available at Kaggle.
The dataset contains 27 categorical and numeric attributes that describe the health condition of horses sample. Three output classes indicate what happened to the horse: died, lived or was submitted to euthanasia.
The "trabalhoDM_5" script contains a Exploratory Data Analysis (EDA), missing values and unnecessary attributes treatment, dataset balancing and different testing prediction models. All of these are essential to the prediction accuracy improvement. The "datadict" file helps to understand the dataset and is available at Kaggle.
RStudio 1.1.383 and a few packages (listed in the script).
- Arthur Silveira
Thanks to Manoela, Data Mining teacher of BI-Master 2017.2 course at PUC-Rio (Brazil), for the script inputs.