Required packages info are given in individual project README
1.Capstone Project: Dogs vs.Cats Redux
-
Built a end-to-end pipeline to classify dog and cat images obtained from Kaggle
-
Achieved 72.8% validation accuracy using 7-layer convolutional neural network in Tensorflow
2.Predicting Boston Housing Prices
-
Built a supervised learning model using Boston housing prices dataset to predict housing prices in the Boston housing area
-
Performed data exploration in Pandas, Numpy and visualized Decision Tree Regressor evaluated in R^2 and optimized the max depth to combat overfitting
3.Creating Customer Segments
-
Used Principle Component Analysis (PCA) to reduce the dimensionality of customers’ annual spending amounts of diverse product categories from the original 6 dimensions to 2 with a combined explained variance 70.81%
-
Identified various customer segments hidden in data using Gaussian-mixture and K-Means clustering algorithm
-
Optimized wholesale distributor’s delivery service based on the customer segments identified
4.Train a SmartCab How to Drive
-
Trained a smartcab to learn an optimal driving strategy for driving on the city roads while obeying traffic rules, avoiding accidents and reaching destinations on time using reinforcement learning
-
Achieved A+ in both safety and reliability ratings in 200 testing trials
5.Student Intervention System
-
Built a student intervention system that allows school officials to identify students who might need early intervention before they fail to graduate, using 30 features such as ‘age’, ‘sex’, and their graduation pass/fail dataset
-
Explored and evaluated different models such as Logistic Regression, Support Vector Machine, Decision Trees and K-Nearest Neighbors using F1 score as the performance metric
-
Achieved 88.44% F1 testing score using SVM tuned by feature normalization and grid-search optimization