Anamoly detection and Prediction

Can anyone explain the difference between anamoly detection, prediction and normal temperature sensing project.???

Anomaly detection - 1.Z score analysis is used for anomaly detection.
2.Variable thresholds.

Prediction - 1.Polynomial Regression is used for ML and prediction.
2.Train the ML model for better outcome.

Temp sensing - 1.To detect the current temperature and alert user when the
sensor data falls out of range.
2.Static thresholds.

The Anomaly Detection & Prediction features are inhouse, custom-built data science models founded on an hybridation of timeseries decomposition models and clustering methods. The basic models have been tweaked, improved using millions of data points, and adapted to web analytics data.

Anamoly Detection
This involves identification of any out-of-ordinary data/behavior in a dataset. One way is to implement a Z-Score Analysis. Here we calculate the range of acceptable values from a sensor, and check for any values that are outside this range. The threshold is varying and not fixed. It can be improved with a longer frame-size. e.g. If fridge is open.
This involves calculation of future values/data in a dataset. One way is to implement a Polynomial Regression Algorithm where use Machine Learning to find a polynomial function to ‘fit’ or describe our available data values. We can then use this function to predict or find values at future timestamps. The function can be improved by training better. There is no threshold since our purpose is not to detect current changes in values but to predict future ones. e.g. Sales of a product.
Normal Sensing Projects
This involes observing sensor values & notifying the user when the values fall out of the specified range. The threshold remains constant. Performance is improved by selecting good threshold values. e.g. Lab temperature detection.
A light-sensor (LDR) will detects the amount of light from morning to noon and the sensor readings rise steadily.

  • Anamoly Detection will be used to detect instances when the amount of light has a sudden increase/decrease (by shining light onto it or by blocking light). It is not triggered by normal light since the threshold keeps changing.
  • Prediction will be used to find what the light levels will be later into the day. So, by 10 AM we may know the light levels at 11 AM.
  • Normal Sensing will notify the user when the light levels are above/below a set threshold. This can occur early morning (low) or at noon (high) with medium values.

Anomaly detection is certain change in value of parameter it can be observed by plotting graph and can be solved using z-score method