I have a data set to predict the sales price for houses. I ran a PCA analysis on the continuous variable without changing or cleaning the data. The analysis suggested that 26 component out of the 31 componnet explained 96% of variability.
So I saved the values of these 26 component. Now I am trying to run a neural netwrok analysis by choosing all the 26 saved component and all the remaining categorical variables as X factor and Sales price as Y. What I realize is that when I increease the NTanH value in the first layer the R-Square keeps increasing. The R-Square increases for NTanH 3 to 10 then reduces at NTanH 15 and then again increases at NTanH 40 etc.
WHY DID THE VALUE DECREASE FOR 15 AND THEN AGAIN START INCREASING??
P.S ignore the the last three predicted sales price column.