Is my GLM model with a binomial distribution correctly implemented?

I've been trying to implement a GLM model using data on the success of a genetic test (yes= successful test; no= unsuccesful test).

> head(dataraw)
   success     pre during season observer
1:      no pre-wet    dry winter   JvD&OK
2:     yes pre-wet    dry winter   JvD&OK
3:      no pre-wet    dry winter   JvD&OK
4:     yes pre-wet    dry winter      JvD
5:     yes pre-wet    dry winter      JvD
6:     yes pre-wet    dry winter      JvD

Four predictor variables are used to explain the ocurrence of the response variable success, being pre (pre-wet or pre-dry), during (wet or dry),season (winter or fall) and observer (up to 10 different observers).

I would like to find which variables are the most important in explaining a succesful test, i.e. success:yes.

I've constructed the models following the code below with and without interactions between the different effects, and have chosen the most parsimonious model following a theoretical approach with AIC values:

m1 <- glm((success) ~ pre , data=dataraw , family=binomial)
summary(m1)
plot(allEffects(m1))
AIC(m1)

m2 <- glm((success) ~ during , data=dataraw , family=binomial)
summary(m2)
plot(allEffects(m2))
AIC(m2)

m3 <- glm((success) ~ season , data=dataraw , family=binomial)
summary(m3)
plot(allEffects(m3))
AIC(m3)

m4 <- glm((success) ~ observer , data=dataraw , family=binomial)
summary(m4)
plot(allEffects(m4))
AIC(m4)

m5 <- glm((success) ~ pre*during , data=dataraw , family=binomial)
summary(m4)
plot(allEffects(m4))
AIC(m4)

etc.

I'm unsure whether I'm following the good approach and if my code is correct, specially since I've seen other people use 1 (for yes) and 0 (for no) when using a binomial distribution. Does that matter? Is my dataset dataraw implemented correctly?

Hope somebody can set me on the right track and I hope this question can be of interest.