The goal of this project is to apply and demonstrate the use of different machine learning methods to real datasets.
We had to choose two datasets within the following constraints :
- several hundreds of lines
- at least 6 attributes (columns), the first being a unique id, separated by commas
- we may use some categorical (non quantitative) features.
- some fields should be correlated
Obvously, we had to tweak the datasets in order to artificially make it possible to apply analysis and visualization techniques on them.
In a first part, we had to work on a dataset and perform supervised learning on it. We decided to work on mobile price classification, and try to find some relation between features of a mobile phone (RAM, number of cores, internal memory...) and its selling price.
We found a dataset that instead of giving the actual price of each phone give a price range indicating how high the price is.
We could perform different analysis ont this dataset and try to visualize our data.
We decided to test several algorithms with the sklearn and xgboost library.
First, we split our data to obtain a train dataset and a test dataset, then we iterate on the classifier to fit them and test them. The scoring is here a R2 score, which is evaluated by cross validation.
The results are the following :
In this section, we decided to work on a dataset containing information about 167 countries, like the inflation, the net income by person and other economical parameters.
After loading the data we checked that there were no missing and duplicated data. And we decided to visualize it with seaborn library.
Our objectives in this unsupervised learning is to decide whether or not a country should receive a Funding for Development Aid.
For this machine learning part, we decided to use the K-Means algorithm from the sklearn library. The objective was to apply a clustering technique on our data to classify them and to know which one of them should receive a Funding for Development Aid.
We applied the algorithm on two datasets, the first one which was the result of a PCA, and the second one which was the result of a MinMax Scaling. The most optimized number of cluster is 3, we used the elbow method and the silouhette method from the sklearn library to find it.
alexandre.lemonnier
alexandre.poignant
sarah.gutierez
victor.simonin