diem-ai / datascience-projects Goto Github PK

• Analyse the data and create new features
  
• Transform text data into Term Frequency - Inverse Document Frequency, select the best feature with f_classif and fit the transformed data to Bayesian algorithm
  
• Transform text data into Word Embedding, select the best feature with f_classif and fit the transformed data to Bayesian algorithm
  
• Word Embedding + Bayes improves 8% accuracy from 86% (baseline) to 94% Meanwhile TF-IDF + Bayes improve 5% accuracy from 86% to 91%
  
• Text Classification with TF-IDF, Word Embedding and Naive Bayes
  

Wine Clustering

• Clustering wine color from their chemical properties. Using KMeans as feature engineering in Classification
  Exploring how hyperparameter tuning and cross validation impact to the improvement
  KMeans - Clustering Method Part 1

Customer Segmentation

• Explore the variation of different customers and predict the segmentation throught their channel products and spending
  Dimensionality Reduction with Principal Component Analysis

Movie Recommendation

• Recommend movies to users using collaborative filtering and content based techinques
• Collaborative filtering: Recommendation list is generated based on the most similar items to a user's already-rated items
• Content based Model: Recommend movies with similar contents : genres, actors, actresses, crew
• Techniques: Data Cleaning, Data Visualization, NLP

Sentiment Mining

• Explore Logistic Regression Classifier with postive/negative/neutral product's reviews
• Extrac topics in customer's review with Laten Dirichlet Allocation (LDA)
• Generate postitive/negative/neutral reviews by implementing Marko Chain Text Generator
• Techniques: Data Visualization, Data Cleaning, Classification, Topic Modeling

Portfolio Investment Optimization

• Collect historical data of 20 stocks of S&P 500 in 5 years
• Use the Principle Component Analysis from Sklearn to structure the eigenvector features of covariance matrix of stocks
• Calculate the weights of each portfolio in PCA components
• Compute the sharpe ratio, annual return and annual volatility of each portfolio
• Techniques: Data scrapping, Data Visualization, Principal Components Analysis