Vikramaditya R. Jakkula & Diane J. Cook Washington State University

Index Introduction MavHome Project Experimentation Environment Temporal Relations Experimentation Process Results Conclusion & Future Work

Smart Homes: Goals Adapt to Needs Cost Effective and Reliable Maximum Comfort and Security 3VJ © 2007

MavHome: Smart Home Project 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 © 20074

Experimentation Environment 1 5VJ © 2007

Experimentation Environment 2 MavHome Environment  MavLab  MavKitchen  MavPad 6VJ © 2007

Experimentation Environment 3 MavLAB Argus Sensor Network  around 100 Sensors.  include Motion, Devices, Light, Pressure, Humidity and more. 7VJ © 2007

What is a temporal relation? Food “Contains” Water or Water “Before” Pills or Food “Meets” Pills or Food “Contains” Water “before” Pills Food Water Pills Time Interval  “It is common to describe scenarios using time intervals rather than time points” - James F. Allen 8VJ © 2007

Why Temporal Relations? Reminder system based on temporal relations. Reminder Assistance If Pills are to be taken “After” Food, we can notice violation of this activity! Anomaly Detection If Cooker is Spoiled should we call emergency or a normal repair? Maintenance If Oven used for Turkey, Is turkey at Home? Temporary Need Analysis Increase prediction accuracy with association rules! Improve Prediction 9VJ © 2007

Temporal Relations (prediction) Before X After During Contains X Overlaps X Overlapped-By Meets X Met-by Starts Started-By Finishes Finished-By Equals 10 VJ © 2007 TEMPORAL RELATION VISUAL DIAGRAM CONSTRAINT USABL E Y After X X During Y YOverlapped-by X Start(X)<Start(Y); End(X)<Start(Y) Start(X)>Start(Y); End(X)<End(Y) Start(X)<Start(Y); Start(Y)<End(X); End(X)<End(Y) Y Met-by X XFinishesY YFinishedbyX XStarts Y YStartedby X X Equals Y Start(Y) = End(X) Start(X)≠Start(Y); End(X) = End (Y) Start(X)≠Start(Y); End(X) = End (Y) Start(X)=Start(Y); End(X)≠End(Y) Start(X)=Start(Y); End(X)=End(Y) X X X X Y Y X X Y Y X X Y Y X X Y Y X X Y Y X X Y Y X X X X Y Y Y Y Y Y

Datasets: Real vs. Synthetic Real Dataset and synthetic datasets consist timestamp of the activity with the activity name and the state it is in. VJ © Real Dataset (Sample): 3/2/ :40:0 AM, (Studio E) E9 OFF 3/2/2003 2:40:0 AM, (Living Room) H9 ON 3/2/2003 2:40:0 AM, (Living Room) H9 OFF 3/2/2003 6:4:0 AM, (Living Room) H9 OFF 3/3/2003 3:43:0 AM, (Studio C) C14 ON 3/3/2003 3:43:0 AM, (Studio C) C15 ON 3/3/2003 3:43:0 AM, (Studio C) C13 ON Synthetic Dataset (Sample): 2/1/ :02:00 AM, off, oven 2/1/ :00:00 AM, on, lamp 2/1/ :11:00 AM, off, thermostat 2/1/ :02:00 PM, off, lamp 2/1/ :35:00 PM, off, cooker 2/1/2006 1:30:00 PM, on, lamp 2/1/2006 2:02:00 PM, off, fan Datasets Parameter Setting No of DaysNo of Events No of Intervals Identified Size of Data Synthetic KB Real KB

Experimentation Current focus on enhanced prediction. Step 1: Temporal Interval analysis and formulation. Step 2: Association rule generation using Weka. Step 3: The run for prediction!

Step 1: Temporal Intervals Process raw data to form temporal intervals. Raw Sensor Data Timestamp Sensor State Sensor ID 3/3/ :18:00 AM OFF E16 3/3/ :23:00 AM ON G12 3/3/ :23:00 AM ON G11 3/3/ :24:00 AM OFF G12 Identify Time Intervals Date Sensor ID Start Time End time. 03/02/2003 G11 01:44:00 01:48:00 03/02/2003 G19 02:57:00 01:48:00 03/02/2003 G13 04:06:00 01:48:00 03/02/2003 G19 04:43:00 01:48:00 Associated Temporal Relations Date time Sensor ID Temporal Relation Sensor ID 3/3/ :00:00 AM G12 DURING E16 3/3/ :00:00 AM E16 BEFORE I14 3/2/ :00:00 AM G11 FINISHESBY G11 4/2/ :00:00 AM J10 STARTSBY J12 Raw Sensor DataInterval Data Temporal Relations Data

Temporal Interval Analyzer Algorithm : Temporal Interval Analyzer Input: data timestamp, event name and state Repeat While [Event && Event + 1 found] Find paired “ON” or “OFF” events in data to determine temporal range. Read next event and find temporal range. Identify relation type between event pair from possible relation types (see Table 1). Record relation type and related data. Increment Event Pointer Loop until End of Input.

Step 2: Association rule generation using Weka Prediction requires strong rules to be formed. Use existing techniques such as Association rule mining to find predictive rules. Use existing tool such as Weka. Use Apriori classifier in Weka for generating best rules with a given support and confidence.

Parameter settings for rules generation using Weka in Real Datasets R UN M INIMUM SUPPORT MINIMUM CONFIDENCE N O OF B EST R ULES F OUND

Parameter settings for rules generation using Weka in Synthetic Datasets R UN M INIMUM SUPPORT MINIMUM CONFIDENCE N O OF B EST R ULES F OUND

Rule Generation Due to small datasets used, we use the top rules generated with a minimum confidence of 0.5 and a minimum support of Confidence level above 0.5 and support above 0.05 could not be used, as they could not result in any viable rules. Sample of best rules observed in real datasets: Activity=C11 Relation=CONTAINS 36 ==> Activity=A14 36 Activity=D15 Relation=FINISHES 32 ==> Activity=D9 32 Activity=D15 Relation=FINISHESBY 32 ==> Activity=D9 32 Activity=C14 Relation=DURING 18 ==> Activity=B9 18

ActiveLeZi: A sequential Predictor Model Useful for prediction of events with previous history. Based on LZ78 text compression algorithm Employs Markov Models to optimally predict the next event in the sequence.

Step 3: Temporal Rules Enhancement to the Prediction. Input: Output of ActiveLezi Predictor a, Best Rules r, Temporal Dataset Repeat If a! = null Repeat Set r1 to the first event in the relation rule If (r1 ==a) Then If (Relation! = “After”) Then Calculate evidence (use Equation 1) If evidence > (Mean + 2 Std. Dev.) is noted Then Make event related to r1 in the best rule as next predictor output; Else *Get next predicted event and look for there temporal relation in the temporal relations database based on the frequency. If again the relation is after Then goto * Until no more “After” relations are found Calculate evidence If evidence > (Mean + 2 Std. Dev.) Then predict; Else Calculate evidence and if evidence > (Mean + 2 Std. Dev.) Then predict this event based on the relation; End if. Until end of rules. End if. Loop until End of Input.

Equation 1: Probability Model P(Z|Y) = |After(Y,Z)| + |During(Y,Z)| + |OverlappedBy(Y,Z)| + |MetBy(Y,Z)| + |Starts(Y,Z)| + |StartedBy(Y,Z)| + |Finishes(Y,Z)| + |FinishedBy(Y,Z)| + |Equals(Y,Z)| / |Y|

Experiment Results Comparing ActiveLezi based prediction with and without temporal rules D ATA S ET A CCURACY %E RROR % R EAL (W ITHOUT R ULES ) 5545 S YNTHETIC (W ITHOUT R ULES ) 6436 R EAL (W ITH R ULES ) 5644 S YNTHETIC (W ITH R ULES ) 6931

Experiment Results % prediction performance improvement in the real data and 7.81% improvement in the synthetic data. Clear indications of larger datasets would lead to drastic improvement in prediction.

Conclusion Unique and new Approach. Real data had 1.86% and synthetic data had 7.81% improvement. Larger datasets would be incorporated and experimented soon.

Future Direction Expansion of the temporal relations by including more temporal relations, such as until, since, next, and so forth, to create a richer collections. Temporal Visualization problems. Pattern analysis for lifestyle and behavior improvements.

Time for Discussions! VJ © Thank You