EarthquakeDamage_Neeraj.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Introduction"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Within the United States alone, earthquakes destroy nearly $4.4B of economic value yearly. Our team will be delving into the 2015 Nepal Earthquake Open Data Portal. This data, which was collected using mobile devices following the Gorkha Earthquake of April 2015, details the level of destruction brought upon more than 200,000 buildings in the area. By utilizing various features reported by Nepali citizens (such as building size, purpose, and construction material), we will construct a machine learning classifier capable of determining the extent to which a building would be damaged. We will also run unsupervised clustering algorithms to search for features that best represent the damage classifications. This project, combined with technologies such as the CV-based city scanners following earthquakes (Ji, 2018) will ultimately provide a better understanding of susceptibility to earthquake-induced damage, valuable information that can be leveraged by city planners."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Data Exploration"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Write about initial filtered data to figure out approaches, results that tell us what to do"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Expected Results"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Using supervised machine learning algorithms we hope to identify which factors affect the level of damage to a building from an earthquake. We’ll compare each of the results by micro averaged F1 score, which will balance precision and recall modified to gauge accuracy for classification into 3 categories (DrivenData). We can also use dimensionality reduction to reduce the number of features from 38 to a more manageable amount by seeing which features are correlated to each other."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Methods"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The data has been collected for us, but we plan to spend a significant amount of time preparing (cleaning, encoding, etc) the data for the model. We will also perform meaningful visualizations to better understand the relationships between our features. The classifier will be trained using supervised learning algorithms such as Support Vector Regressors (Asim, 2018), Logistic Regression, and Hybrid Neural Networks (Asim, 2018) and unsupervised learning algorithms such as KMeans clustering. The hyperparameters will be tuned using the GridSearchCV process. The model performances will be compared using our test set and the one that ​generalizes t​ he best will be chosen.\n",
    "\n",
    "- Multiple Logistic regression\n",
    " - Support vector regression\n",
    " - Kmeans\n",
    " - DBSCAN\n",
    " - PCA\n",
    " - LDA\n",
    " - Cross validation\n",
    " - Hybrid Neural Network\n",
    " - Decision Tree\n",
    "  https://www.datasciencecentral.com/profiles/blogs/decision-tree-vs-random-forest-vs-boosted-trees-explained\n",
    " - Random Forest Decision Tree\n",
    " - XGBoost Decision Tree\n",
    " (add plots under section option)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## DBSCAN"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Tried DBSCAN, but with the high dimensionality of the data (39 features with one-hot encoding) it was very difficult to get greater than 0 clusters unless the epsilon parameter was very high. In order to achieve some clusters the epsilon value had to be greater than 40 even with just a minpts of 3. Anything below that epsilon would result in 0 clusters.\n",
    "\n",
    "After trying some more variants of eps, a trial using an eps of 42, minpts of 3 and euclidian distance as the metric, we were able to get 3 clusters"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## KMEANS"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Tried KMEANS while setting the number of n_clusters to 3 in hopes of clustering them close to damage grade. Just like with DBSCAN it seems KMEANS clustering does not perform well either with high dimensional data. KMEANS was able to provide an output of each datapoint to one of 3 clusters but after comparing all variations of cluster assignments to damage grade, the maximum accuracy we got was 33.5%."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Results"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Synthesize information from Methods"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Discussion"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This project can benefit architects, engineers, and city planners by using the classification model to extrapolate and predict types of buildings that are likely to suffer from earthquake damage. Buildings with attributes similar to those that were more damaged can be reinforced. Both the visualization and classification models can be used in conjunction with earthquake prediction research (Rouet-Leduc, 2017) to provide advance humanitarian aid so buildings can be reinforced to take significantly less damage.\n",
    "\n",
    "Future plans, etc."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# References"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Add references"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}