Imbalanced Machine Learning For Fraud Detection

cost-sensitive xgboost,RusBoost,Smote,EasyEnsemble,cost-sensitive logistic regression

Introduction Data with imbalanced target class occurs frequently in several domians such as credit card Fraud Detection ,insurance claim prediction, email spam detection, anomaly detection, outlier detection etc. Financial instituions loose millions of dollars every year to fraudulent financial transactions. It is important that these institutions are able to identify... [Read More]

Automatic Machine Learning

automl in Python and R

Automated Machine Learning Automated Machine Learning (AutoML) has increased greatly the efficiency of building machine learning models. AutoML achieves this by automating in some applications data pre-processing ,feature engineering, feature extraction , feature selection and hyper-parameter tuning when building machine learning models. AutoML has also reduced the expertise in academic... [Read More]

Receiver Operating Characteristic AUC Packages in R

Comparison of Varoius Packages

ROC curves and AUC values are common evaluation metric for binary classification models. Although there are some criticism of it especially its’s appropritatenes in evaluating models built with imbalanced data, they still remain the most popular evaluation metric for binary classification models. In the case of highly imbalanced classification, the... [Read More]

Machine Learning Interpretability With R

Introduction The two main packages in R for machine learning interpretability is the iml and DALEX. H2o package also has built in functions to perform some interpretability such as partial dependence plots. DALEX and iml are model agnostic as such can be used to explain several supervised machine learning models... [Read More]

Machine Learning Interpretability

Permutation Importance, Partial Dependence Plots, SHAP values, LIME, lightgbm,Variable Importance

Introduction Machine learning algorithms are often said to be black-box models in that there is not a good idea of how the model is arriving at predictions.This has often hindered adopting machine learning models in certain industires where interpretation is key. Examples of such areas include financial institutions who are... [Read More]