Reaction time analysis using subsets of features

Introduction

The number of car accidents is increasing constranly, one of the main causes is due to the use of cell phones. The reaction time to external events can be affected by the use of cel phones

Pre-processing

We are using the dataset reaction.time from the package Using R (1). The dataset was divided 70 % for training and 30 % for testing purposes, using in both cases random sampling without replacements. We used Shiny, UsingR and caret packages

library(shiny)
library(UsingR)
library(caret)
data(reaction.time)

Pre-processing

process<-function(features) {
  reaction<-subset(reaction.time,select=c(features,"time"))
  set.seed(123)
  nsample<-sample(1:nrow(reaction), 18)
  train_set<-reaction[-nsample,]
  test_set<-reaction[nsample,]
  modFit<-train(time~.,method="rpart",data=train_set)
  fit<-modFit$finalModel
  pred<-predict(modFit,subset(reaction,select=features))
  mse<-mean((pred-test_set$time)^2)
  return(list(mse=mse, data=head(test_set),model=fit))
}

Method

We are usinging partinioning for classification. Each time the user select or unselect an attribute the algorithm is retrained with the update attributes and calculates MSE (Mean Square error) on the testing data.

mse<-process(c("age","gender","control"))[1]
test<-process(c("age","gender","control"))[2]
model<-process(c("age","gender","control"))[3]

Results

For this presentation we used the three attributes (age,gender and control) in order to determine the MSE (Mean Square Error).

Results

For this presentation we used the three attributes (age,gender and control) in order to determine the MSE (Mean Square Error).

test
$data
     age gender control  time
18 16-24      M       T 1.352
47   25+      F       C 1.523
24   25+      M       C 1.595
51   25+      M       T 1.445
53   25+      M       T 1.517
3  16-24      M       T 1.512

Results

For this presentation we used the three attributes (age,gender and control) in order to determine the MSE (Mean Square Error).

model
$model
n= 42 

node), split, n, deviance, yval
      * denotes terminal node

1) root 42 0.29030 1.415  
  2) controlT< 0.5 15 0.10790 1.364 *
  3) controlT>=0.5 27 0.12120 1.444  
    6) age25+< 0.5 11 0.03181 1.391 *
    7) age25+>=0.5 16 0.03802 1.480 *

Results

For this presentation we used the three attributes (age,gender and control) in order to determine the MSE (Mean Square Error).

mse
$mse
[1] 0.0115

Conclusion

A Specific combination of attributes gives more meaningful results than using a single attributeor even using the whole set of features.

Bibliography

  1. from package UsingR, version 0.1-18. License: GPL (>= 2)