Transcription
MACHINE LEARNING APPROACHES USING MATLABP REMA N A N DA I N D I C, P H . D .DEPARTMENT OF ELECTRICAL ENGINEERING
OUTLINE INTRODUCTION DIFFERENT MACHINE LEARNING APPROACHES DISCUSSION
OUTLINE INTRODUCTION DIFFERENT MACHINE LEARNING APPROACHES DISCUSSION
INTRODUCTION What is Machine Learning ? Machine Learning is a field of study that gives computers the ability to βlearnβwithout being explicitly programmed Prediction ClassificationSamuel AL, IBM J. Research & Development, 1959, vol. 3 (3), 210-229
INTRODUCTION Too many books spoil the curiosity Start with Andrew Ng, Machine Learning, Stanford University available on YouTubeSome Statistics & Programming Knowledge Helps !
INTRODUCTION Machine Learning with MATLABhttps://commons.wikimedia.org/wiki/File:Man Driving Car Cartoon ts.htmlMachine Learning Driving School
INTRODUCTIONORS Webinar ngs.phpMinuteFeature ExtractionStatistical orMachine lDataPreprocessingFeature ExtractionFeature Selection
INTRODUCTIONText EditorWorkspace30MinuteFeature ExtractionCommand windowDataStatistical orMachine LearningModelsPreprocessingFeature ExtractionFeature Selection
OUTLINE INTRODUCTION DIFFERENT MACHINE LEARNING APPROACHES DISCUSSION
Statistical vs. Machine Learning ModelsPurpose:Statistical models are used for inference (To find association between features andan outcome). Results should be interpretable.Machine Learning models are used for prediction (Use features that can predict anoutcome). Results may not be interpretable.
Statistical vs. Machine Learning ModelsAssociation vs. PredictionHealthy Individual2000100000Philips Actiwatch 2N 24123Individual with depression20001000001ππΌ π ππΌ πΆπππΌπ ππππΌπΆ πππΌ ππππΌ2ΰ·© π ππΌ πππΌSensitivity & Specificity3
LEARNING APPROACHES Supervised LearningLearning a relationship between features and the outcome using a training set Unsupervised LearningLearning underlying structures in features
LEARNING APPROACHES Supervised Learning Linear Regression Logistic Regression Support Vector Machine Artificial Neural Network . .
LEARNING APPROACHES Unsupervised LearningClustering Principal Component Analysis Independent Component Analysis Singular Value Decomposition .
LEARNING APPROACHES Do machines actually βlearnβ ?N 24ππΌ π ππΌ πΆ
LEARNING APPROACHES Do machines actually βlearnβ ?N 24ΰ·ͺπΌ π 1 ππΌ(π 1)π π 1 πΰ·ͺπΌ π 2 ππΌ π 2π π 2 π .ΰ·ͺπΌ π 24 ππΌ(π 24)π π 24 πΰ·ͺπΌ π ππΌ πΆπ2πΈ Οπππ 1
LEARNING APPROACHES Do machines actually βlearnβ ?How do we find minimum E ?N 24mC 00120.1120.892120.50.05ΰ·ͺπΌ π ππΌ πΆπ
LEARNING APPROACHES Do machines actually βlearnβ ?How do we find minimum E ?N 24- Gradient DescentEby Louis Augustin Cauchy in 1847ΰ·ͺπΌ π ππΌ πΆπLinear Regressionΰ·© π ππΌ πππΌmC
LEARNING APPROACHES Do machines actually βlearnβ ?Classification of High Risk (n 43) vs. Low Risk (n 95)0 Low Risk, 1 High RiskP 1π VI1 π (π ππΌ π )MeanVarianceSkewnessKurtosis Linear RegressionPowerPeriod1π1 π (π΄ πΉ π΅ )10FLogistic RegressionAccuracy 73%
LEARNING APPROACHES How to implement in MATLAB ?Step 1: Create an excel sheet with featureswith class assignments
LEARNING APPROACHES How to implement in MATLAB ?Step 2: Open MATLAB and drag theexcel file to workspace
LEARNING APPROACHES How to implement in MATLAB ?Step 3: Click Import Selection andimport data
LEARNING APPROACHES How to implement in MATLAB ?Step 4: Features are in workspace and ready
LEARNING APPROACHES How to implement in MATLAB ?Step 5: Go to Apps,-click classification learner,- -select Logistic Regressionfrom Model Type-click New Session,-select from Workspace
LEARNING APPROACHES How to implement in MATLAB ?Step 6: Set 10 fold Cross validation- Start the session
LEARNING APPROACHES NONLINEAR FEATURESCravings DetectionCarreiro, S, Chintha KK, Shrestha S, Chapman B, Smelson D, Indic P. Wearable sensor based detection of stress and craving in patients during treatment for substance usedisorder: A mixed methods pilot study. Drug and Alcohol Dependence. 2020, 107929.
LEARNING APPROACHES2 : No Stress3: StressStress Detection AlgorithmSloke Shrestha
LEARNING APPROACHESStress Detection Algorithm2 : No Stress3: Stress
SUMMARYoooooIdentification of FeaturesDevelop MATLAB code for feature extractionSet up databaseAssist with experiment protocol and data analysisMachine LearningDataPreprocessingStatistical orMachine LearningModelsFeature ExtractionFeature Selection
SUMMARYo Ready to go features and Machine Learning ModelsCHIAfeaturesStatistical orMachine LearningModelsfast & efficientDataPreprocessingFeature ExtractionFeature Selection
THANK YOUCurrent Students:Sloke Shrestha, UGMohammed Alenazi, GraduatePravitha Ramanand, PhD, PostdocFormer Students:Joshua Stapp, GraduateApurupa Amperayani (PhD Student, Arizona State University)Jonathan Wells (PhD Student, UT Austin)Pallavi AtluriKeerthi Chintha (Data Scientist, Wabtec Corporation)Selorm Darkey (Business Intelligent Analyst, Taylor Solutions)
THANK YOUSBIR: RAE (Realize, Analyze, Engage) - A digital biomarkerbased detection and intervention system for stress andcarvings during recovery from substance abuse disorders.PIs: M. Reinhardt, S. Carreiro, P. IndicSTARs AwardThe University of Texas SystemP. Indic (PI, UT Tyler)Design of a wearable sensor system and associated algorithm to tracksuicidal ideation from movement variability and develop a novelobjective marker of suicidal ideation and behavior risk in veterans.Clinical Science Research and Development Grant (approved forfunding),P. Indic (site PI, UT-Tyler)E.G. Smith (Project PI, VA)P. Salvatore (Investigator, Harvard University)Design of a wearable biosensor sensor system with wirelessnetwork for the remote detection of life threatening events inneonates.National Science Foundation Smart & ConnectedHealth GrantP. Indic (Lead PI, UT-Tyler)D. Paydarfar (Co PI, UT-Austin)H. Wang (Co PI, UMass Dartmouth)Y. Kim (Co PI, UMass Dartmouth)Pre-VentNational Institute Of Health GrantP. Indic (Analytical Core PI, UT-Tyler)N. Ambal (PI, Univ. of Alabama, Birmingham)Wearable system for the detection of addictionP. Indic (PI, UT-Tyler)M. Reinhart (PI, ContinueYou, LLCS. Carriero, (PI. Univ. of Mass. Med. School)
DISCUSSION
How to implement in MATLAB ? Step 1: Create an excel sheet with features with class assignments. LEARNING APPROACHES How to implement in MATLAB ? Step 2: Open MATLAB and drag the excel file to workspace. LEARNING APPROACHES How to implement in MATLAB ? Step 3: Click Import Selection an