System and method for property condition analysis

1 . A method, comprising:

determining, by a computing system, a set of measurements depicting a property, the set of measurements including imagery data corresponding to a time window;

for each measurement, extracting, by the computing system, a feature value vector from the measurement using a feature extractor, the feature value vector being indicative of a visual feature of the property, wherein each feature value vector is extracted by:

determining a semantic segmentation mask based on the measurement using a semantic segmentation model;

segmenting property feature pixels from the measurement based on the semantic segmentation mask;

identifying property feature pixels based on the semantic segmentation mask; and

extracting the feature value vector, based only on the identified property feature pixels, using the feature extractor;

for each feature value vector:

determining, by the computing system, a scaling factor indicating a weight of the visual feature in determining a condition of the property; and

applying, by the computing system, the scaling factor to obtain a scaled feature value vector;

determining, by the computing system, a set of attribute values based on the scaled feature value vector of each measurement, using a machine learning model;

determining, by the computing system, a condition score for each end of the time window based on the set of attribute values, using a condition scoring model;

determining, by the computing system, a property analysis indicating a change in a property condition based on the condition score; and

increasing or decreasing an insurance premium based on the change in the property condition and the condition score,

wherein the attribute values comprise semantic attribute values including at least one of discoloration, tarp presence, material degradation, missing material, sealing, or defects.

2 . The method of claim 1 , wherein the machine learning model comprises an RNN.

3 . The method of claim 1 , wherein the condition score comprises a roof condition score.

4 . The method of claim 1 , wherein extracting the feature value vector from the measurement comprises extracting the feature value vector based on pixels of the measurement.

5 . The method of claim 4 , wherein the pixels comprise a subset of the measurement, identified using the semantic segmentation model.

6 . The method of claim 1 , wherein the time window is predetermined.

7 . The method of claim 1 , wherein the property analysis comprises a timeseries analysis.

8 . The method of claim 1 , wherein the condition scoring model is trained by:

determining training data for a training property, the training data comprising:

a set of training measurements for the training property; and

a training data metric determined based on historical claim data for the training property; and

training the condition scoring model to predict the training data metric based on a set of training attribute values determined based on the set of training measurements.

9 . The method of claim 8 , wherein the training data metric comprises a claim loss metric.

10 . The method for claim 8 , wherein the training data is used to train the machine learning model.

11 . The method of claim 1 , wherein the set of measurements comprises measurements from different measurement modalities.

12 . The method of claim 1 , wherein the set of measurements comprises measurements from different measurement perspectives.

13 . The method of claim 1 , wherein the set of attribute values are further determined based on at least one of historical weather data or historical hazard data.

14 . The method of claim 1 , wherein the set of attribute values are further determined based on time data.

15 . The method of claim 1 , wherein the set of measurements comprises remote exterior imagery.