The invention discloses a milling cutter wear prediction method and a state recognition method. The wear prediction method comprises the following steps that firstly, wavelet noise reduction processing is performed on milling vibration data, feature extraction is performed on vibration signals from three aspects including time domain, frequency domain and time domain, after an initial feature vector set is obtained, a correlation coefficient method is used for calculating the correlation between feature vectors and wear amount, and an optimal feature vector set is obtained by screening; then,an average relative error predicted by a least squares support vector machine is defined as a fitness function of an adaptive step size cuckoo search algorithm, and by searching for a nest position, input parameters of the least squares support vector machine are optimized; finally, the wear amount is predicted by using the optimal least square support vector machine. Through comparison with two other hybrid intelligent algorithms, the superiority of an ASCS- LSSVR algorithm is verified.