Statistical-Learning Group Project

Rijin Baby & Angelina Khatiwada,
MSc Data Science and Economics, UNIMI

June, 2021

Dataset: In-vehicle coupon recommendation

Data was collected via a survey on Amazon Mechanical Turk. The survey describes different driving scenarios including the destination, current time, weather, passenger, etc., and then ask the person whether he will accept the coupon if he is the driver (Y is a binary response variable).

Project Description:

Project Kaggle Link: https://www.kaggle.com/rijinbaby/analysis-in-vehicle-coupon-recommendation

Exloratory Analysis and Data Preparation:

• Basic exploratory analysis
• Missing & unique values check
• Dropping irrelevant variables
• Creating new variables
• Missing imputation using KNN approach
• Plotting variables

Modelling - Part 1: Classification (Supervised Learning)

• Logistic Regression with all the parameters
• Multicollinearity check
• Stepwise selection models (both directions, forward, backward)
• Train-test split and Cross-Validation
• Confusion matrix and Sensitivity/Specificity Trade-off
• Penalised models (Lasso and Elastic Net)
• Linear Discriminant Analysis
• Random Forest (parameters tuning, model evaluation)
• Boosting

Modelling - Part 2: Clustering (Unsupervised Learning)

• Gower distance for mixed type data
• K-medoids Clustering
• Agglomerative clustering (different linkages)
• Clustering Evaluation with Rand Index
• PCA

Modelling - Part 3: ANN (Supervised Learning)

• Train-test split
• Neural Network using Keras for Binary Classification
• Network parameters: Sigmoid/Hard_sigmoid activation function, Binary Cross-Entropy loss function, etc.
• Cross-Validation on train set
• Accuracy on test set

R Packages used

skimr, readr, plyr, dplyr, purrr, VIM, ggplot2, plotly, caret, grid, gridExtra, pROC, MASS, class, gmodels, randomForest, car, ClusterR, cluster, gbm, Rtsne, glmnet, dendextend, fossil, leaps

Python Libraries used

tensorflow, keras, scikit-learn, numpy, pandas, matplotlib

To display html reports in a browser, please use the following links:

• https://rawcdn.githack.com/rijinbaby/Statistical-Learning/main/Group_16_ANN_final.html
• https://rawcdn.githack.com/rijinbaby/Statistical-Learning/main/Group_16_Supervised_Unsupervised_final.html