Predictive_Analytics_project_fall_2018 (Oct -Dec-2018)
- Built a credit card anomaly detection system using the CRISP- DM process.
- Handled imbalanced dataset using Random Undersampling, Random Oversampling, Random Oversampling using SMOTE analysis.
- Performed exploratory data analysis for data visualization and feature selection.
- Clustered Data into fraudulent and non-fraudulent transactions using dimensionality reduction technique with t-distributed Stochastic Neighbour Embedding.
- Built predictive models using Logistic Regression (93.65%), K-Nearest Neighbours (99.98%), Support Vector Machine (99.91%), Stochastic Gradient Descent (94.26%).
- Used Ensemble modelling technique and performed Voting Ensemble (95.36%), bagging with Random Forest Classifier (100%), boosting with XGBoost (98.34%) and AdaBoost (96.89%).
- Performed model evaluation and comparison of the models using Area under the ROC Curve (Accuracy) and time for speed of detection.
Technologies used โ Python 3.6, Scikit-Learn, Pandas, NumPy, Matplotlib, Seaborn