📝 Welcome to Tomy’s personal blog!

The content of this blog is consist of Data Science projects built using Python and R programming languages. Several domains covered are as follows:

  • data collection,
  • data visualization,
  • machine learning,
  • deep learning, and
  • many more!

Visit tags for complete categories of content, or search content that matches with your taste 🌟

This site is powered by fastpages.

Posts

  • MNIST Digit Classifier using PyTorch

    A simple workflow on how to build a multilayer perceptron to classify MNIST handwritten digits using PyTorch. We define a custom Dataset class to load and preprocess the input data. The neural network architecture is built using a sequential layer, just like the Keras framework. We define the training and testing loop manually using Python for-loop. The final model is evaluated using a standard accuracy measure.

    Aug 27, 2021

  • Convolutional Neural Network (CNN) for Gender Determination by Morphometry of Eyes

    Workflow on building a CNN model in Keras to classify whether a scanned image of an eye is male or female. We utilize flow_from_dataframe since the label is provided on a separate csv file. Data augmentation technique is applied to further improve model performance and generalization. During model fitting, callbacks such as ModelCheckpoint and EarlyStopping are used.

    Aug 5, 2021

  • Interpretable Loan Approval Pipeline Model

    A simple workflow to demonstrate on how to utilize scikit-learn pipeline to classify whether a loan application should be granted or not. Grid search over the pipeline can be applied to evaluate combination of preprocessing and classifier hyperparameters. The best pipeline is interpreted using the Local Interpretable Model-Agnostic Explanations (LIME) method.

    Apr 15, 2021

  • Sales Forecasting using Prophet

    A comprehensive workflow to forecast software sales using the prophet library, provided by Facebook. This post breakdowns each components of time series, including trend, seasonality, holiday effect, and external regressors. The model is then evaluated using RMSLE by expanding window cross validation. Hyperparameter tuning is done using the grid search algorithm.

    Jan 15, 2021

  • Explainable Heart Disease Classifier with Shapley Additive Explanations (SHAP)

    A simple workflow to classify whether a patient has a heart disease or not using a Logistic Regression model. SHAP explainer is used to further explain the model decision via several plots, such as SHAP force, summary, dependence, and decision plot.

    Dec 21, 2020

  • Survival Analysis for Predictive Maintenance

    An implementation of survival analysis model for predicting the survival probability of a machine over time. We discuss both theoretical and mathematical concepts of survival analysis and its implementation using the pysurvival package. We explore several survival analysis approaches, such as Kaplan-Meier, Cox Proportional Hazard, Linear, and Neural Multi-Task Logistic Regression (MTLR) models. Models are compared using Concordance index, Integrated Brier Score, and prediction error. The best-fitted model is then used to predict the survival function over time and rank individuals based on their risk score.

    Nov 27, 2020

  • Time Series Clustering using Hierarchical-Based Clustering Method

    Curated material for 'Time Series Clustering using Hierarchical-Based Clustering Method' in R programming language. The primary objective of this material is to provide a comprehensive implementation of grouping taxi pick-up areas based on a similar total monthly booking (univariate) pattern. This post covers the time-series data preprocessing, introducing Dynamic Time Warping (DTW) as a distance matrix, two approaches of hierarchical clustering (Agglomerative and Divisive), and ways to evaluate clustering algorithm using agglomerative/divisive coefficient, elbow, and silhouette method.

    Nov 4, 2020

  • Breast Cancer Wisconsin Classification

    A simple machine learning workflow to classify whether a mass is classified as malignant or benign cancer, based on the clinical measurement of cell nucleus. Four models are compared to achieve best F1 score: logistic regression, support vector, decision tree, and random forest classifier. Each models are tuned using grid search cross-validation and threshold tuning. This post is dedicated for a final project of Indonesia AI Mentorship - Machine Learning Program.

    Oct 26, 2020

  • Hotel Booking Cancellation Prediction using Random Forest

    Machine learning workflow to predict the possibility of hotel booking cancellation using Random Forest model. Features are selected using the Predictive Power Score (PPS) method. Several techniques of model improvement are done using downsampling, random search cross-validation, and threshold tuning. Models are evaluated using the confusion matrix and ROC curve.

    May 29, 2020

  • Convolutional Neural Network (CNN) for Casting Product Quality Inspection

    Workflow on building a CNN model in Keras to classify whether a top-view image of casted submersible pump impeller is defect or ok. Data augmentation technique is applied to further improve model performance and generalization.

    May 2, 2020

  • •
  • 1
  • 2