We used an isobar model to study the K+Σ− photoproduction reaction on a neutron target with focus on the resonance region. In order to achieve a reasonable agreement with the data, we included spin-3/2 and spin-5/2 nucleon resonances in a consistent formalism together with a Δ resonance and two kaon resonances on top of the Born terms. The free parameters of the model were adjusted to the data from the CLAS and LEPS Collaborations on differential cross sections and photon beam asymmetry. The cornerstone of this analysis was an upgrade of the fitting method. Previously, we used only the plain χ2 minimization, which could not prevent us from overfitting the data. We, therefore, introduced a regularization method, the least absolute selection shrinkage operator (LASSO), which, together with information criteria, restricts the number of nonzero parameters and prevents us from overfitting the data. In our analysis, we arrived at two models, fit M, whose parameters were fitted with the MINUIT code only, and fit L, where we used the more advanced LASSO. Both models describe the data in a similar way and we observe only slight differences in the dσ/dΩ data description at very forward angles where the fit M is flat whereas the fit L produces two broad peaks, and in the photon beam asymmetries above 2 GeV at backward kaon angles where the fit M produces a bump. Surprisingly, no hyperon resonances are needed for the correct data description in these models. On the other hand, the N(1720)3/2+ nucleon resonance was found to be very important in both models.