Can someone explain variance and bias of data in natural language (not mathematically) I read many tutorial but until now i don't understand what the variance and bias means.
So can someone explain me what these words means clearly, not in mathematical language. 
 A: I will oversimplify quite a bit, there are a lot more subtleties than presented, but I hope this gives a decent 1st order overview.
Variance
Variance measures the range of values we expect to see if we repeatedly measure the same thing. Each time we measure something we get a slightly different answer, even in a situation where the scenario is expected to be stable. The randomness arises from many sources but variance measures how much the final answer moves around the mean of the repeated measures. 
It is a measure of how confident we can be of the value reported based on our measurement protocol is an accurate reflection of that protocol. It measures the noise in the data collection/measurement protocol.
Bias
Bias is measuring the difference between the mean value of our measure and what it should be (so we need to have some way of defining that).  It is the difference between our final answer and the reference answer. If you have a so called 'gold standard' reference then this would be the 'true' answer. It is a measure of how confident we can be of the value reported based on our measurement protocol is an accurate reflection of the reference value. It measures the noise in the model.
Relation between the two
These two are unrelated to each other in the same way as mean and standard deviation are unrelated. 
If you have a very noisy unbiased method (high variance, low bias) then you have a noisy data but on average it reflects the expected underlying situation. Individual measures cannot be trusted, the model may be useful for describing an underlying process but will not be useful for prediction.
If you have a low noise biased method (low variance, high bias) then you have a repeatable data but on average it does not reflect the expected underlying situation. Individual data measures can be trusted to be accurate but the model cannot.
If you have a very noisy biased method (high variance, high bias) then you have a noisy data and on average it does not reflect the expected underlying situation. Individual measures cannot be trusted, nor the model.
If you have a low noise unbiased method (low variance, low bias) then you have a repeatable data and on average it reflects the expected underlying situation. Individual data measures and the model can be trusted to be accurate.
