concept target in category AI

appears as: target, target, The target, targets
Interpretable AI

This is an excerpt from Manning's book Interpretable AI.

  • Classification: This is a class of problems where the target variable y is discrete and bounded. For example, predicting if an email is spam or not is a classification type of learning problem.
  • The objective of the linear regression learning algorithm is to determine the weights that accurately predict the target variable for all patients in the training set. There are two techniques that can be applied here:

    Figure 2.22 shows the effect of each of the 6 blood test measurements on the target. As an exercise, observe the effects that features like total cholesterol, LDL, HDL and glaucoma have on the progression of diabetes. What can you say about the impact of higher LDL values (or bad cholesterol) on the target variable? Why does higher total cholesterol have lower impact on the target variable? In order to answer these questions, let’s look at a few patient cases with very high cholesterol values. The code snippet below will help you zoom into those patients.

    print(df_data[(df_data['Total Cholesterol'] > 0.15) &
                  (df_data['LDL'] > 0.19)])

    If you execute the code above, you will see only 1 patient out of 442 that has a total cholesterol reading greater than 0.15 and an LDL reading greater than 0.19. The fasting glucose level for this patient one year out (the target variable) seems to be 84 which is in the normal range. This could explain why in Figure 2.22 we are seeing a very large negative effect for total cholesterol on the target variable for a range that is greater than 0.15. The negative effect of total cholesterol seems to be greater than the positive effect the bad LDL cholesterol seems to have on the target. The confidence interval seems much wider in these range of values. The model may have overfit on this one outlier patient record and so, we should not read too much into these effects. By observing these effects, we can identify cases or range of values where the model is sure of the prediction and cases where there is high uncertainty. For high uncertainty cases, we can go back to the diagnostics center to collect more patient data so that we have a representative sample.

    Zero to AI: A non-technical, hype-free guide to prospering in the AI era

    This is an excerpt from Manning's book Zero to AI: A non-technical, hype-free guide to prospering in the AI era.

    Eng: Sure, the list of features seems like a good start. And what target do you need from the model? Just the expected property price?