Predictor: An Alternative Approach to Uncertain Inference in Expert Systems

An a l t e r n a t i v e approach to uncer ta in inference in exper t systems is desc r i bed which might be r e g a r d e d as a s y n t h e s i s o f t e c h n i q u e s f r o m au tomat i c i n d u c t i o n and mathematical s t a t i s t i c s . I t u t i l i s e s a type o f p a t t e r n match ing in which compar isons are made between new cases (as yet u n c l a s s i f i e d ) and a database of past cases ( i n which the outcome is known). The method uses a s tepw ise approach in w h i c h , a t each s t a g e , t h a t e v i d e n c e v a r i a b l e p r o v i d i n g t h e g r e a t e s t add i t i ona l d i s c r i m i n a t i n g power between c lasses (above t h a t a l r eady ob ta ined) i s u t i l i s e d . I t avo ids r e l y i n g on the assumpt ion o f c o n d i t i o n a l independence. A r u d i m e n t a r y sys tem (PREDICTOR) w h i c h opera tes acco rd ing to these p r i n c i p l e s has been w r i t t e n . Various adaptations to deal w i th missing and u n c e r t a i n evidence are d e s c r i b e d , as are add i t i ona l features such as a window, a f a c i l i t y fo r focusing d i s c r i m i n a t i o n on a subset of classes and a mod i f i ca t ion to deal wi th sub jec t ive data. II AN ALTERNATIVE APPROACH Another approach to the problem of uncer ta in i n f e r e n c e in exper t systems might l o o s e l y be desc r i bed as a s y n t h e s i s of techn iques f rom au tomat i c i n d u c t i o n and mathematical s t a t i s t i c s . I t u t i l i s e s data on past cases in o rder to make pred ic t ions or c l a s s i f i c a t i o n judgements about new ones. I n t h i s r e s p e c t , i t i s s i m i l a r t o ideas o n a u t o m a t i c i n d u c t i o n i n t h e A I f i e l d ( e . g . M i c h a l s k i , 1983) and a l so to a l a rge f a m i l y of s t a t i s t i c a l t echn iques . However, i t d i f f e r s from automatic induct ion in that the l a t t e r is r e a l l y a l o g i ca l procedure w h i c h , by i t s very n a t u r e , canno t h a n d l e u n c e r t a i n t y . I t d i f f e r s f r o m orthodox s t a t i s t i c a l techniques in that no model is constructed fo r p r e d i c t i v e purposes i n s t e a d each new case is matched aga ins t past data on a subset of v a r i a b l e s , by a process about to be d e s c r i b e d . Other d i f f e r e n c e s are that the system is capable of p r o v i d i n g more of the customary features of expert systems than are usual ly found i n t r a d i t i o n a l s t a t i s t i c a l approaches.