To predict the perceived pleasantness of an odour based on some molecular features in a regression task.

R version used for this project : R version 3.6.1 (2019-07-05)

Plateform used : jupyler lab

Library used that need to be install : ggplot2, caret, leaps, tree, xgboost, Matrix

DataSet in a csv file : "training_data.csv"

Presentation of the data set (pleasantness to be predicted in fuction of intensity and some molecular features).

Cleaning of the data.

Performance of multilinear regression : with forward regresstion, Lasso L1 and L2 regularization and PCA.

Performance of nonlinear methods : descision tree, prunned tree and boosting tree.

All the methods are tested by computing the resulting MSE.

The MSE is evaluated on a test set as the method are computed using a training set.

The methods implemented can be used to predict valence pleasantness using a dataset with the same predictors as in the initial file "training_data.csv".

example : Predict valence pleasantness using the predictors of a given file (test_data) and the boosting tree method of this project, the results are written in a csv file named 'monprojet.csv' :

Data.prediction<- read.csv(file.path("..","data","test_data.csv"))

lookup <- c("low" = 1, "high" = 0)

Data.prediction$Intensity <- lookup[Data.prediction$Intensity]

Data<-Data.prediction[, colnames(Data.prediction)%in%colnames(plea.train.x)]

Data.x <- xgb.DMatrix( data.matrix(Data), missing = NA) pred<-predict(boost.pleas, Data.x)

result<- data.frame('Id'=Data.prediction$Id, 'VALENCE.PLEASANTNESS' = pred)

write.csv(x=result, 'monprojet.csv', row.names = FALSE )

example : Predict valence pleasantness using the predictors of a given file (test_data) and the pruuned tree method of this project, the results are written in a csv file named 'tree.csv' :

Data.kaggle1<- read.csv(file.path("..","data","test_data.csv")) lookup <- c("low" = 1, "high" = 0) Data.kaggle1$Intensity <- lookup[Data.kaggle1$Intensity]

#Data.kaggle.x<-Data.kaggle1[, colnames(Data.kaggle1)%in%colnames(data.train.x)]

pred<-predict(final.tree, Data.kaggle1)

result<- data.frame('Id'=Data.kaggle1$Id, 'VALENCE.PLEASANTNESS' = pred)

write.csv(x=result, 'tree.csv', row.names = FALSE )

example : Predict valence pleasantness using the predictors of a given file (test_data) and the tree method of this project, the results are written in a csv file named 'tree.csv' :

Data.kaggle1<- read.csv(file.path("..","data","test_data.csv")) lookup <- c("low" = 1, "high" = 0) Data.kaggle1$Intensity <- lookup[Data.kaggle1$Intensity]

#Data.kaggle.x<-Data.kaggle1[, colnames(Data.kaggle1)%in%colnames(data.train.x)]

pred<-predict(p_tree, Data.kaggle.x)

result<- data.frame('Id'=Data.kaggle1$Id, 'VALENCE.PLEASANTNESS' = pred)

write.csv(x=result, 'tree.csv', row.names = FALSE )