Styleguide

Portfolio

In order to expand and improve my skills in Machine Learning and Data Science I have explored multiple public datasets, which are described in this page. Most of the code can be seen in my GitHub.
My latest work is related to Football Data Science, area in which I am currently focusing on.

Check my work!

Football Domain
One of my personal areas of interest is Football, as such, some data analysis and modeling were performed using football (event/tracking) data, namely:
- Creation of a Pitch Control Model, which reflects the probability of the team getting the ball possession at a given field position
- Explore different Pitch Control variants, focused on Ball Possession Retention, Vertical Game, of Game by the Flanks
- Computing the value of a given action, using SPADL structure data, based on the probability of a given action will end in goal
- Predicting if a game action will lead to a goal or not, using Machine Learning
- Plots of different game features, like a game move, shots frame and players passing and shooting accuracy
- Analysing the importance of features (like game stats and individual players) in the matches results
- Creation of an Expected Goal Model, using only distance to goal as input to a SVM classifier
- Players Heat Map, using tracking data
- Predict a substitute player rating, based on the game statistics until the moment he enters in the game
- Determination of the best/most suitable position for a given player (using FIFA 20 dataset)
- Implementation of a Genetic Algorithm for finding the best set of players to buy under a given budget, for given positions, considering different criteria, as, for example, the player age (using FIFA 20 dataset)
- Capture Player playing style, using Non Negative Matrix Factorization
- Player detection and Shirt number identification from game footages, using Deep Learning
- Capture Team passing patterns, using K-Means applied to passing sequences
- Also, a Tableau dashboard of Manchester City season 2018/2019 and beginning of 2019/2020 is available at https://public.tableau.com/profile/daniel.azevedo#!/vizhome/Dashboard_ManchesterCity_Analysis/Home_DashboardIn this dashboard one can see the team performance in terms of results (home and away), scores, game stats, player individual stats, players importance on the game result, and much more. The data was scraped from web and saved in BigQuery with an ETL job.
- All this work can bee seen in:https://football-data-science.herokuapp.com https://github.com/danielazevedo/Football-Analytics

Next, some more generic Data Science and Machine Learning projects are described, applied over different domains.
Each project is based on a different dataset, with a specific goal in mind, nevertheless, in all of them there is an initial Exploratory Data Analysis (EDA) followed by the application of specific Machine Learning models depending on the goal to be achieved.
Some of the projects have a Demo option available.

Note: Click on the project title to see more details regarding the work done with that specific dataset.

Regression
- House Prices Prediction Dataset
  - Missing Data handling, categorical encoding and standard normalization were applied
  - Features were selected based on their correlation, a XGBoost model was trained to predict the house prices
  - https://github.com/danielazevedo/Kaggle-competitions/tree/master/House%20Prices%20-%20Advanced%20Regression%20Techniques
- London Bike Sharing Dataset
  Demo
  - EDA was focused on the understanding of the relationship between the different features
  - Different techniques for feature engineering and features selection were explored
  - https://github.com/danielazevedo/Kaggle-competitions/blob/master/London%20bike%20sharing/EDA_bikes.ipynb
Classification
- Credit Card Fraud Detection Dataset
  Demo
  - Class balancing, robust normalization were applied. Features were selected based on correlation and statistical testing
  - Two models were compared in order to identify fraudulent credit card transactions: Logistic Regression and SVM
  - https://github.com/danielazevedo/Kaggle-competitions/blob/master/CreditCardFraud/CreditCard%20-%20EDA.ipynb
- Fake News Dataset
  - Classes balancing, Missing data handling and text cleaning were applied
  - Naive Bayes classifier was trained to detect fake news based on their title or text corpus
  - https://github.com/danielazevedo/Kaggle-competitions/blob/master/Fake_News_Detector/fake_news_detector.ipynb
Computer Vision
- Digit Recognizer Dataset:
  Demo
  - Data reshaping, data normalization and categorization were applied
  - A CNN was trained to identify the digits appearing in the different images
  - https://github.com/danielazevedo/Kaggle-competitions/blob/master/Digit%20Recognizer/digit-recognizer-using-a-cnn.ipynb
NLP
- Cipher Decryption
  Demo
  - A substitution cipher is decrypted, based on a Language Model represented by a Bi-gram
  - The algorithm used for decryption is a Evolutionary Algorithm
  - https://github.com/danielazevedo/Kaggle-competitions/blob/master/CipherDecryption_NLP/Cipher_DA.ipynb
- SMS Spam Detector
  Demo
  - Feature engineering and features importance were analysed
  - Two methods for feature extraction were explored: CountVectorizer and TfidfVectorizer
  - https://github.com/danielazevedo/Kaggle-competitions/blob/master/SMS_Spam_Detector_NLP/spamDetector.ipynb
- Phrases Sentiment Analysis:
  Demo
  - Features extracted using CountVectorizer. Words stemming was also explored
  - Logistic Regression was trained to identify the phrases positive/negative connotation.
  - https://github.com/danielazevedo/Kaggle-competitions/blob/master/SentimentAnalysis_NLP/sentiment_analysis.ipynb

House Prices Prediction Dataset

London Bike Sharing Dataset

Credit Card Fraud Detection Dataset

Fake News Dataset

Digit Recognizer Dataset:

Cipher Decryption

SMS Spam Detector

Phrases Sentiment Analysis: