Systems and methods for determining states using run-length encoders, binarized bins, and k-means cluster models

1 . A non-transitory processor-readable medium storing code representing instructions to be executed by one or more processors, the instructions comprising code to cause the one or more processors to:

receive (1) past interaction data associated with (a) a mobile device and (b) a user, (2) current interaction data associated with the mobile device and the user, and (3) past state data associated with the user;

generate, based on the past interaction data, an expected time window indication;

generate, based on the current interaction data, an inferred time window indication;

generate a consistency metric based on the inferred time window indication and the expected time window indication;

generate a disruption metric based on the current interaction data and the expected time window indication;

predict a current state associated with the user based on the past state data, the consistency metric, and the disruption metric;

predict a first action based on the current state; and

provide the current state and an indication of the first action as input to a machine learning model to predict a future state, the machine learning model being configured based on a feedback signal that is generated based on (a) the past state data and (b) a second action determined before the first action.

2 . The non-transitory processor-readable medium of claim 1 , wherein the current state is a first current state, and the instructions further include code to cause the one or more processors to:

predict a second current state associated with the user;

based on a difference between the second current state and the first current state, cause a state request to be sent to the mobile device;

receive an indication of a third state from the mobile device in response to the state request;

determine a directionality metric associated with the first action based on the third state; and

modify the first action to produce a third action based on the directionality metric.

3 . The non-transitory processor-readable medium of claim 1 , wherein the instructions further include code to cause the one or more processors to generate an anomaly alert based on the current state.

4 . The non-transitory processor-readable medium of claim 1 , wherein:

the consistency metric is associated with a time-variant overlap percentage.

5 . The non-transitory processor-readable medium of claim 1 , wherein the past interaction data and the current interaction data each includes a sequence of bins, each bin from the sequence of bins being associated with a predefined time interval.

6 . The non-transitory processor-readable medium of claim 1 , wherein the code to cause the one or more processors to generate the inferred time window indication includes code to cause the one or more processors to:

provide a sequence of bins included in the current interaction data as input to a run-length encoding function to produce at least one inactivity period indication;

generate, for each 24-hour period from at least one 24-hour period associated with the current interaction data, the inferred time window indication based on an inactivity period indication from the at least one inactivity period indication, the inactivity period indication specifying a longer duration than a remaining inactivity period indication from the at least one inactivity period indication; and

generate the inferred time window indication based on the inferred time window indication generated for each 24-hour period from the at least one 24-hour period.

7 . The non-transitory processor-readable medium of claim 1 , wherein the code to cause the one or more processors to generate the expected time window indication includes code to cause the one or more processors to:

generate, based on (1) the past interaction data included in a sequence of bins and (2) a predefined window length, an activity metric;

provide the past interaction data and the activity metric as input to a k-means clustering model to sort the past interaction data to produce an activity cluster and an inactivity cluster;

binarize the activity cluster to produce a binarized activity cluster;

binarize the inactivity cluster to produce a binarized inactivity cluster;

generate a sequence of binarized bin data based on the binarized activity cluster, the binarized inactivity cluster, and the sequence of bins; and

generate, for the predefined window length, the expected time window indication based on the sequence of binarized bin data.

8 . The non-transitory processor-readable medium of claim 7 , wherein the inactivity cluster has a lower activity metric centroid than the activity cluster.

9 . The non-transitory processor-readable medium of claim 1 , wherein the past interaction data and the past state data are each associated with longitudinal data.

10 . The non-transitory processor-readable medium of claim 1 , wherein the consistency metric includes at least one of an indication of a longest period of inactivity, an indication of a regularity of the current interaction data, or an overlap percentage associated with (1) the current interaction data and (2) the expected time window indication.

11 . The non-transitory processor-readable medium of claim 1 , wherein the current interaction data includes time data and an application change indication.

12 . A method, comprising:

receiving, at a processor, longitudinal interaction data associated with a mobile device and including time data;

sorting, via the processor, the longitudinal interaction data into a plurality of bins based on the time data, each bin from the plurality of bins being associated with a time interval;

providing, via the processor, the plurality of bins as input to a run-length encoder to produce at least one inactivity period indication;

generating, via the processor, an inferred period indication based on the at least one inactivity period indication;

generating, via the processor and based on the plurality of bins, a plurality of activity metrics;

providing, via the processor, the plurality of activity metrics and the plurality of bins as input to a k-means clustering model to produce an inactivity cluster;

generating, via the processor, an expected period indication based on the inactivity cluster;

generating, via the processor, a first characteristic value based on the inferred period indication and the expected period indication;

providing, via the processor, the first characteristic value as input to a first machine learning model to determine a first series of clinical actions; and

predicting, via the processor, a future state by providing the first characteristic value and the first series of clinical actions as input to a second machine learning model trained based on a feedback signal that is generated based on (a) a second characteristic value generated before the first characteristic value and (b) a second series of clinical actions determined before the first series of clinical actions.

13 . The method of claim 12 , further comprising generating an anomaly alert based on the first characteristic value.

14 . The method of claim 12 , wherein the longitudinal interaction data includes time data and an application change indication.

15 . The method of claim 12 , wherein:

the sorting includes sorting the plurality of bins to produce the inactivity cluster and an activity cluster; and

the inactivity cluster has a lower activity metric centroid than the activity cluster.

16 . The method of claim 12 , wherein the first characteristic value includes at least one of an indication of a longest period of inactivity, an indication of a regularity of the longitudinal interaction data, or an overlap percentage associated with (1) the inferred period indication and (2) the expected period indication.

17 . The method of claim 12 , wherein the generating the first characteristic value includes generating the first characteristic value based on circumstance data.