Prediction Mapping Using Gis Data And Advanced Ml Algorithms

eXtreme Gradient Boosting, K Nearest Neighbour, Naïve Bayes, Random Forest for Prediction Geo-Hazards and Air pollution
What you'll learn
Two applications of susceptibility prediction mapping in GIS, 1) Landslides prediction maps 2) Ambient air pollution prediction maps
Step by step analysis of ML algorithms for classification: eXtreme Gradient Boosting (XGBoost) K nearest neighbour (KNN) Naïve Bayes (NB) Random forest (RF)
Run classification based algorithms with training data model accuracy, Kappa index, variables importance, sensitivity analysis of explanatory and response data
Hyper-parameter optimization procedure and application
Model accuracy test and validation using; confusion matrix and results validation using AUC value under ROC plot
Produce prediction maps using Raster and vector dataset
Requirements
No prior knowledge in programming needed
Basic knowledge in R studio environment
Basic knowledge in GIS and QGIS
Basic knowledge about man made and natural hazards
Description
In this course, four machine learning supervised classification based techniques used with remote sensing and geospatial resources data to predict two different types of applications: Project 1: Data of Multi-labeled target prediction via multi-label classification (multi class problem). Target (Y) that has 3 labeled classes (instead of Numbers): Names, description, ordinal value (small, large, X-large)..Multiple output maps. Like:Increase specific type of species in certain areas and its relationship with surrounding conditions.Air pollution limits prediction (Good, moderate, unhealthy, Hazardous..)Complex diseases types: potential risk factors and their effects on the disease are investigated to identify risk factors that can be used to develop prevention or intervention strategies.Course application: Prediction of concentration of particulate matter of less than 10 µm diameter (PM10)This project was published as research articles using similar materials and with major part of analysis (with slight modification to the code). "Demystifying uncertainty in PM10 susceptibility mapping using variable drop-off in extreme-gradient boosting (XGB) and random forest (RF) algorithms" in Environmental Science and Pollution Research journal.Project 2: Data of Binary labeled target prediction. Target with 2 classes: Yes and No, Slides and No slide, Happened –Not happened, Contaminated- Clean.Flooded areas and it contribution factors like topographic and climate data.Climate change related consequences and its dragging factors like urban heat islands and it relationship with land uses.Oil spills: polluted and non polluted.Course application: Landslide susceptibility mapping in prone area.If you are previously enrolled in my previous course using ANN, then you have the chance to compare the outcomes, as we used the same landslide data here.Eventually, all the measured data (training and testing), were used to produce the prediction map to be used in further GIS analysis or directly to be presented to decision makers or writing research article in SCI journals.This course considered the most advanced, in terms of analysis models and output maps that successfully invested in the (1) machine learning algorithm and geospatial domains; (2) free available data of remote sensing in data scarce environment.
Overview
Section 1: Introduction and Course Content : Get to know what will we talk about!
Lecture 1 Course contents
Lecture 2 Course applications: Landslide and Air pollution prediction analysis
Lecture 3 Projects data, study areas and applications extent
Lecture 4 Expected outcomes: What will we achieve together!
Section 2: Practical summary about the classification based machine learning algorithms!
Lecture 5 CARET package in R
Lecture 6 Hyperparameters optimization (model tuning) in machine learning
Lecture 7 eXtreeme Gradient Boosting (XGBoost) classifier machine learning
Lecture 8 K- nearest neighbors (KNN) classifier machine learning
Lecture 9 Naive Bayes (NB) classifier machine learning
Lecture 10 Ensemble Random forest (RF) classifier machine learning
Lecture 11 Selection of training and testing data concept
Lecture 12 Current computer and software's specifications that used in the course
Section 3: Project 1: PM10 prediction mapping : Data record pre-processing and data entry
Lecture 13 PM10 readings pre-processing and input data preparation in Excel
Lecture 14 Allocate the air monitoring stations and record data entry in QGIS
Lecture 15 PM10 readings conversion to WHO limits in QGIS
Section 4: Project 1 PM10 prediction mapping : Input data-frame processing and production
Lecture 16 Preparation of PM10 prediction remote sensing variables data-list
Lecture 17 Landsat 8 imagery download
Lecture 18 Visualization of downloaded Landsat 8 images
Lecture 19 Processing of Landsat 8 bands and indices in R
Lecture 20 Processing of Land Surface Temperature (LST) in R
Lecture 21 Processing of average monthly and annual Landsat 8 bands and indices in R
Lecture 22 Processing and production of road networks variable in QGIS
Lecture 23 Preparation of input dataframe (target and conditioning factors) in QGIS
Lecture 24 Finalize input variables and convert it to CSV format file in QGIS for modeling
Section 5: Project 1 PM10 prediction mapping : modeling of advanced ML classifiers in R
Lecture 25 XGBoost algorithm: Data entry and visualization in R
Lecture 26 XGBoost algorithm: Run of train default function
Lecture 27 XGBoost algorithm: Hyper-parameter optimization and plot (model tuning)
Lecture 28 XGBoost algorithm: AUC value of ROC plot
Lecture 29 XGBoost algorithm: Fit optimized model using all inventory observations
Lecture 30 XGBoost algorithm: Conversion to dataframe and scaling of Raster images
Lecture 31 XGBoost algorithm: Probability prediction maps production
Lecture 32 XGBoost algorithm: Classification prediction maps production
Lecture 33 NB algorithm: ggplot of linearity between target and independents and variables
Lecture 34 NB algorithm: Run of train default function
Lecture 35 NB algorithm: Hyper-parameter optimization, AUC of ROC plot & normalized Rasters
Lecture 36 NB algorithm: Probability and classification prediction maps production
Lecture 37 KNN algorithm: Run of train function and hyper-parameter optimized models
Lecture 38 KNN algorithm: AUC of ROC and probability and classification prediction maps
Lecture 39 RF algorithm: Data entry and train function using Grid search tuning
Lecture 40 RF algorithm: train function using Random search tuning and AUC of ROC
Lecture 41 RF algorithm: Scaling and conversion of raster images to dataframe
Lecture 42 RF algorithm: Probability prediction map
Lecture 43 RF algorithm: Classification prediction map
Lecture 44 Summary and Visualization of 4 algorithms prediction resultant maps in QGIS
Section 6: Project 2 Landslide : Create training and testing data in QGIS
Lecture 45 Adding my developed tools to QGIS processing library
Lecture 46 Create Land Cover map (convert string observations to numeric) in QGIS
Lecture 47 Run the tools Step 1
Lecture 48 Run the tools Step 2
Lecture 49 Run the tools Step 3
Section 7: Project 2 Landslide prediction mapping : pre-processing training data in Excel
Lecture 50 Excel work step 1
Lecture 51 Excel work step 2
Section 8: Project 2 Landslide prediction mapping : modeling of advanced ML classifiers
Lecture 52 XGBoost algorithm : Training and testing data entry in R
Lecture 53 XGBoost algorithm : Run train function using default settings
Lecture 54 XGBoost algorithm: Hyper-parameter optimization (model tuning) and pairs plot
Lecture 55 XGBoost algorithm: AUC of ROC plot and important technical error
Lecture 56 XGBoost algorithm: Run optimized model and probability prediction maps
Lecture 57 XGBoost algorithm: Classification prediction map production
Lecture 58 KNN algorithm : Data entry and visualization of target and other variables
Lecture 59 KNN algorithm: Run of train function and hyper-parameter optimized models
Lecture 60 KNN algorithm: AUC of ROC plot and technical issues with data entry
Lecture 61 KNN algorithm: probability prediction maps
Lecture 62 KNN algorithm: classification prediction map
Lecture 63 NB algorithm: Training data entry and visualization of variables
Lecture 64 NB algorithm: Train function and Hyper-parameters and AUC of ROC plot
Lecture 65 NB algorithm: Probability and classification prediction maps production
Lecture 66 RF algorithm: Data entry of training data variables
Lecture 67 RF algorithm: default train function and Hyper-parameter and AUC of ROC plot
Lecture 68 RF algorithm: Probability and classification prediction maps
Lecture 69 Summary and Visualization of 4 algorithms prediction maps in QGIS
Section 9: Projects Conclusion and main remarks of the presented course
Lecture 70 Summary: Let us sum up everything and recap what we discussed earlier!
All students, researchers and professionals that interested in using data mining with GIS Data,All students, researchers and professionals that work on: Health[viruses susceptibility, noise maps, Epidemic expansions, Infectious Disease, Famine,All students, researchers and professionals that work on: Hazards[ flooding, landslides, geological based, drought, air pollution..]
Screenshots

#