milogreen.github.io
Turning big data into big insights, and coffee into code.
Project maintained by mkgreen Hosted on GitHub Pages — Theme by mattgraham
email me: green.milok@gmail.com
Find me on LinkedIn
Data Science Projects
Here are some of my best Data Science Projects. I have explored various machine-learning algorithms for different real-world datasets. Please feel free to contact me to learn more about my experience working on these projects.
Diagnosis of breast cancer using a logistic classifier
Skills used: Python, Pandas, SKlearn, Matplotlib
Project Objective: Identification of the type of Breast Cancer for quicker diagnosis. This assists professionals in the medical field to take appropriate measures to accurately diagnose, treat, and save lives.
Quantifiable result: Types of tumors were successfully classified with 96% accuracy by using K-means algorithm.
- Used logistic regression to identify a tumor as malignant or benign based on various attributes
- Classified tumors as benign or malignant by studying patterns in measured attributes of those tumors
- Used Logistic regression classifier & optimized the accuracy by using the ROC curve
- Explored a machine-learning approach to medical diagnosis
Amazon Fine Food Analysis using NLP
Skills used: Python, Pandas, SKlearn, TfidVectorizer
Project Objective: Given a review, determine whether the review is positive or negative based on Amazon foods.
Quantifiable result: A rating of 4 or 5 could be considered a positive review. A review of 1 or 2 could be considered negative. A review of 3 is neutral and ignored. This is an approximate way of determining the polarity (positivity/negativity)
- Given a review, it is determined whether the review is positive or negative.
- Used NLP for this approach.
- A review of 1 or 2 could be considered negative. A review of 3 is neutral and ignored.
Identifying symptoms of orthopedic patients as normal or abnormal
Skills used: Python, Pandas, SKlearn, Matplotlib,KNN,NB
Project Objective: In this project, the data provided multiple instances of orthopedic parameters and their classification as either Normal or Abnormal. I implemented K Nearest Neighbor to classify and diagnose the patients.
Quantifiable result: Successfully classified the orthopedic parameters as either Normal or Abnormal with an accuracy of 81%.
- Used the K Nearest Neighbours algorithm to classify a patient’s condition as normal or abnormal based on various orthopedic parameters
- Compared predictive performance by fitting a Naive Bayes model to the data
- Selected best model based on train and test performance
TalkingData Project on Bagging and Boosting ensemble model:
Skills used: Python, Pandas, SKlearn, Matplotlib, XGboost Classifier, BaggingClassifier
Project Objective: Predict fraud by using the data gathered from features associated with clicks, such as IP address, operating system, device type, time of click, etc.
Quantifiable result: Successful classification of people who did and did not download the app after watching an advertisement with an accuracy of 97%.
- Used the XGboost Classifier algorithm to classify the People who downloaded the app after watching the advertisement and identify the click fraud
- Compared predictive performance by fitting a BaggingClassifier model to the data
- Selected best model based on train and test performance
Prediction of user interest using bank data
Skills used: Python, Pandas, SKlearn, Matplotlib
Project Objective: Using the real-world data of a Portuguese banking institution, predict if a client will subscribe to a term deposit (variable y).
Quantifiable result:
- The classification goal is to predict if the client will subscribe a term deposit (variable y).
Examining the effect of environmental factors and weather on demand of Bike rentals
Skills used: Python, Pandas, SKlearn, Matplotlib
Project Objective: Predicting Bike rental demand based on weather and seasonal factors in advance to take appropriate measures which finally will result in bike utilization.
Quantifiable result: Successfully able to predict the Bike rental demand resulting in 94% accuracy.
- Used Random Forest Regressor to predict the number of bikes rented in the city of Seoul
- The data had quite a few categorical variables which were encoded for use in the model
- Encoded categorical variables to numeric using Sklearn due to the presence of many string columns
- Cross-validation for validating the training data and model fit.
- Fit a Random Forest Regressor with high prediction accuracy through iteration