LATENT CLASS ɆɈȾȿɅɂ ɁА ɋȿȽɆȿɇɌɂɊАɇȿ ɇА ɉАɁАɊА
ɫɬ.аɫɢɫɬɟɧɬ Ⱦɟяɧ Ʌаɡаɪɨв, ɫɬ.аɫɢɫɬɟɧɬ Аɬаɧаɫ Ʌɭɢɡɨв
LATENT CLASS MODELS FOR MARKET SEGMENTATION
Deyan Lazarov, Atanas Luizov
Ɋɟɡюɦɟ: В ɞɨɤɥаɞа ɟ ɩɪɟɞɫɬаɜɟɧ ɧаɤɪаɬɤɨ ɟɞɢɧ ɨɬ ɧаɣ-ɡɧаɱɢɦɢɬɟ ɫɬаɬɢɫɬɢɱɟɫɤɢ
ɦɨɞɟɥɢ – Latent Class. ɉɪɢ ɡаɞаɱɢɬɟ ɡа ɫɟɝɦɟɧɬɢɪаɧɟ ɧа ɩаɡаɪа, ɢɡɫɥɟɞɨɜаɬɟɥɢɬɟ ɫа
ɭɫɬаɧɨɜɢɥɢ, ɱɟ ɢɡɩɨɥɡɜаɧɟɬɨ Latent Class ɦɨɞɟɥɢ ɩɪɢɬɟɠаɜа ɝɨɥɟɦɢ ɩɪɟɞɢɦɫɬɜа ɜ
ɫɪаɜɧɟɧɢɟ ɫ ɬɪаɞɢɰɢɨɧɧɢɬɟ ɩɨɞɯɨɞɢ ɡа ɤɥаɫɢɮɢɰɢɪаɧɟ.
Abstract: In this paper briefly is present one of the most significant statistical model – Latent
Class. In marketing segmentation tasks, researchers are realizing that the use of Latent Class
models cans yield powerful improvements over traditional approaches to cluster.
Key words: segmentation, cluster analysis, Latent Class models, finite mixture models
Ɉɬ ɧɚɱɚɥɨɬɨ ɧɚ 60-ɬɟ ɝɨɞɢɧɢ ɫɟɝɦɟɧɬɢɪɚɧɟɬɨ ɫɟ ɪɚɡɝɥɟɠɞɚ ɤɚɬɨ ɤɥɸɱɨɜɚ ɤɨɧɰɟɩɰɢɹ ɜ
ɦɚɪɤɟɬɢɧɝɚ. Ⱥɤɨ ɟɞɢɧ ɩɚɡɚɪ ɫɟ ɪɚɡɞɟɥɢ ɧɚ ɱɚɫɬɢ (ɫɟɝɦɟɧɬɢ) ɫ ɪɚɡɥɢɱɧɢ ɩɪɟɞɩɨɱɢɬɚɧɢɹ ɢ ɜ
ɪɟɡɭɥɬɚɬ ɧɚ ɬɨɜɚ ɫɬɨɤɢɬɟ ɢ ɭɫɥɭɝɢɬɟ ɫɟ ɩɪɢɫɩɨɫɨɛɹɬ ɤɴɦ ɩɨɬɪɟɛɧɨɫɬɢɬɟ ɜ ɪɚɡɥɢɱɧɢɬɟ
ɫɟɝɦɟɧɬɢ ɳɟ ɛɴɞɟ ɧɚɦɚɥɟɧɚ ɞɢɫɬɚɧɰɢɹɬɚ ɦɟɠɞɭ ɩɪɟɞɥɨɠɟɧɢɟɬɨ ɧɚ ɨɪɝɚɧɢɡɚɰɢɹɬɚ ɢ
ɢɡɢɫɤɜɚɧɢɹɬɚ ɧɚ ɩɨɬɪɟɛɢɬɟɥɢɬɟ. Ɍɟɡɢ ɞɟɣɫɬɜɢɹ ɳɟ ɞɨɜɟɞɚɬ ɞɨ ɩɨɞɨɛɪɹɜɚɧɟ ɧɚ
ɤɨɧɤɭɪɟɧɬɧɚɬɚ ɩɨɡɢɰɢɹ. ɂɞɟɧɬɢɮɢɰɢɪɚɧɟɬɨ ɧɚ ɤɥɚɫɬɴɪɢ ɢɥɢ ɩɚɡɚɪɧɢ ɫɟɝɦɟɧɬɢ ɧɚ
ɨɫɧɨɜɚɬɚ ɧɚ ɢɡɜɚɞɤɚ ɨɬ ɪɟɫɩɨɧɞɟɧɬɢ, ɨɩɢɫɚɧɢ ɨɬ ɫɴɨɬɜɟɬɟɧ ɛɪɨɣ ɡɧɚɱɢɦɢ ɩɪɨɦɟɧɥɢɜɢ,
ɦɨɠɟ ɞɚ ɛɴɞɟ ɧɚɩɪɚɜɟɧɨ ɩɨ ɪɚɡɥɢɱɧɢ ɧɚɱɢɧɢ. Ɋɚɡɜɢɬɢɟɬɨ ɧɚ ɢɧɮɨɪɦɚɰɢɨɧɧɢɬɟ
ɬɟɯɧɨɥɨɝɢɢ ɩɪɚɜɢ ɜɴɡɦɨɠɧɨ ɩɪɢɥɚɝɚɧɟɬɨ ɧɚ ɦɨɞɟɥɢ ɡɚ ɫɟɝɦɟɧɬɢɪɚɧɟ, ɩɪɢɬɟɠɚɜɚɳɢ
ɝɨɥɟɦɢ ɩɪɟɞɢɦɫɬɜɚ ɜ ɫɪɚɜɧɟɧɢɟ ɫ ɬɪɚɞɢɰɢɨɧɧɢɬɟ ɩɨɞɯɨɞɢ.
ɋɴɳɧɨɫɬ ɧɚ Latent Class (LC) ɦɨɞɟɥɢ
Latent Class ɦɨɞɟɥɢɬɟ ɫɟ ɨɬɧɚɫɹɬ ɤɴɦ ɤɥɚɫɚ ɩɪɨɰɟɞɭɪɢ ɡɚ ɫɩɪɚɜɹɧɟ ɫ ɯɟɬɟɪɨɝɟɧɧɨɫɬɬɚ ɧɚ
ɞɚɧɧɢɬɟ. ɉɪɢɟɦɚ ɫɟ, ɱɟ ɞɚɧɧɢɬɟ ɫɚ ɫɴɜɤɭɩɧɨɫɬ ɨɬ ɯɨɦɨɝɟɧɧɢ ɩɨɞɝɪɭɩɢ, ɤɨɢɬɨ ɫɟ ɧɚɪɢɱɚɬ
ɟɥɟɦɟɧɬɢ ɧɚ ɫɴɜɤɭɩɧɨɫɬɬɚ. Ɍɟɡɢ ɦɨɞɟɥɢ ɫɚ ɢɡɤɥɸɱɢɬɟɥɧɨ ɩɨɥɟɡɧɢ ɡɚ ɢɡɜɴɪɲɜɚɧɟ ɧɚ
ɤɥɚɫɢɮɢɤɚɰɢɹ ɢ ɩɪɟɞɫɬɚɜɥɹɜɚɬ ɦɨɞɟɥɟɧ ɩɨɞɯɨɞ ɡɚ ɫɟɝɦɟɧɬɢɪɚɧɟ.