Vikramaditya Jakkula Washington State University First International Workshop on Smart Homes for Tele-Health
Project Unique Focus on entire home House perceives and acts Sensors Controllers for devices Connections to the mobile user and Internet Unified project incorporating varied AI techniques, cross disciplinary with mobile computing, databases, multimedia, and others. VJ ©
Minimal Sequential Patterns Using “ED” Given an input stream S of event occurrences O, ED: Partitions S into Maximal Episodes, Pmax. Creates Itemsets, I, from the Maximal Episodes. Creates a Candidate Significant Episode, C, for each Itemset I, and computes one or more Significance Values, V, for each Candidate. Identifies Significant Episodes by evaluating the Significance Values of the candidates. VJ ©
Decision Making using ProPHeT ProPHeT is the main controlling component of the system. It uses data filtered through Episode Discovery (ED) to create a Hierarchical Hidden Markov Model (HHMM). HHMM represents a user model that includes all of the episodes (e.g., entering a room, watching TV, sitting in a chair and listening to music, and so forth) that a person performs in the environment. VJ ©
6
MavHome Environment MavLab MavKitchen MavPad 7 VJ © 2007
MavHome Smart Apartment The evaluation environment is a student apartment with a deployed Argus and X-10 network. There are over 150 sensors deployed in the MavPad that include light, temperature, humidity, and switches.
Monitoring Health by Detecting Drifts and Outliers for a Smart Environment Inhabitant help us gain information about different types of drifts and outliers that are part of the inhabitant’s lifestyle anomalies in inhabitants health such as blood pressure, pulse and temperature values. Gives information about sudden changes observed in inhabitants health Identification of Lifestyle Behavior Patterns with Prediction of the Happiness of an Inhabitant in a Smart Home Identify correlations among everyday activity in smart home. we use the machine learning technique, the k-nearest neighbor algorithm, to predict the state of wellness the inhabitant will experience on the following day.we use the machine learning technique, the k-nearest neighbor algorithm, to predict the state of wellness the inhabitant will experience on the following day. simple sensor network in smart home can be used to detect lifestyle patternsShow that a simple sensor network in smart home can be used to detect lifestyle patterns
Basic overall goal is to build a forecasting system for healthcare system in smart home. Used 40 days data for training and 20 days data for testing. Experiment 1: Use support vector machine algorithm found in Weka to predict whether health trends are increasing, decreasing or constant. Experiment 2: Use automated forecasting tools, such as Phicast, which enables us to use Box-Jenkins method for forecasting the range of health data values.
We are looking to prediction an increasing, decreasing or constant trend over time. Data values range is very small and close to each other. (Look at the figure). In a 2-D plot let us try to classify the trend. Which technique do you think is better to use here? SVM! Oh yes!
Inst No.ActualPredictedError 12:decrease No 21:increase No 31:increase No 42:decrease No 52:decrease No 62:decrease No 71:increase No 81:increase No 91:increase No 102:decrease No 112:decrease No 121:increase No 132:decrease1:increaseYes 141:increase No 152:decrease No 162:decrease No 172:decrease No 182:decrease No 191:increase2:decreaseYes 202:decrease No Predicted systolic trend using test set DatasetCorrectly classified Instances Incorrectly Classified Instances Mean Absolute Error Test Set Cross Validation Plot of Predicted systolic trend [test set]
InstNo.ActualPredictedError 12:decrease1:increaseYes 21:increase No 31:increase No 42:decrease1:increaseYes 52:decrease1:increaseYes 63:constant1:increaseYes 71:increase No 81:increase No 91:increase No 102:decrease1:increaseYes 112:decrease1:increaseYes 123:constant1:increaseYes 131:increase2:decreaseYes 141:increase No 152:decrease1:increaseYes 161:increase No 172:decrease1:increaseYes 182:decrease1:increaseYes 192:decrease1:increaseYes 201:increase No Predicted diastolic trend using test set Plot of Predicted diastolic trend [test set] DatasetCorrectly classified Instances Incorrectly Classified Instances Mean Absolute Error Test Set Cross Validation
Instance No ActualPredictedError 12:decrease No 21:increase No 31:increase No 42:decrease No 52:decrease No 62:decrease No 71:increase No 81:increase No 91:increase No 102:decrease No 112:decrease No 121:increase No 131:increase2:decreaseYes 141:increase No 152:decrease No 162:decrease No 172:decrease No 182:decrease No 191:increase No 202:decrease No Predicted pulse trend using test set Plot of Predicted pulse trend [test set] DatasetCorrectly classified Instances Incorrectly Classified Instances Mean Absolute Error Test Set Cross Validation
Box-Jenkins forecasting models are based on statistical concepts and principles. Box-Jenkins forecasting models regarded to be the efficient forecasting technique. Box-Jenkins forecasting provides some of the most accurate short-term forecasts. Other tools such as SAS, SPSS, Demeter needed more data for running similar experiment, where as Phicast had a successful run.
Instance # Actual Recorded Systolic Lower Predicte d Value Upper Predicte d Value Error [Yes/No] No No No No Yes Yes Yes Yes No No No No No Yes Yes No No No Yes No Yes No of Correctly Classified No of Incorrectly ClassifiedPercent Accuracy Test Set13862%
Instance # Actual Recorded Diastolic Lower Predicte d Value Upper Predicte d Value Error [Yes/No] Yes No Yes No No No No No No Yes No No No Yes Yes Yes Yes No No No Yes No of Correctly Classified No of Incorrectly ClassifiedPercent Accuracy Test Set13862%
Instanc e # Actual Recorde d Pulse Lower Predicte d Value Upper Predicte d Value Error [Yes/No ] No No No No No No No No No Yes No No No No Yes Yes No No No Yes Yes No of Correctly Classified No of Incorrectly ClassifiedPercent Accuracy Test Set16576%
Predicting time series data is difficult Continue performance on more larger datasets. We observe that the prediction models act as useful components to the health care system in smart homes. Future work would include improving the prediction, collecting more data over time. Anomaly detection based prediction for health care system and activities associated with healthcare.
Thank You