Chapter 2 – Literature Review

2.1 Introduction
Autism spectrum disorderis a neurobiological disease that impacts cognitive and social
development. Autistic people often have trouble interacting with others and communicating, and
they also have narrow interests and routine routines. At present, the identification of Autism
Spectrum Disorder is a significant problem in the field of advanced illness detection.[29]

‫ﻣﻘﺪﻣﺔ‬

‫ ﻏﺎﻟﺒًﺎ ﻣﺎ ﻳﻮﺍﺟﻪ ﺍﻟﻤﺼﺎﺑﻮﻥ ﺑﺎﻟﺘﻮﺣﺪ‬.‫ﺍﺿﻄﺮﺍﺏ ﻃﻴﻒ ﺍﻟﺘﻮﺣﺪ ﻫﻮ ﻣﺮﺽ ﻋﺼﺒﻲ ﺑﻴﻮﻟﻮﺟﻲ ﻳﺆﺛﺮ ﻋﻠﻰ ﺍﻟﻨﻤﻮ ﺍﻟﻤﻌﺮﻓﻲ ﻭﺍﻻﺟﺘﻤﺎﻋﻲ‬
‫ ﻳﻌﺪ‬،‫ ﻓﻲ ﺍﻟﻮﻗﺖ ﺍﻟﺤﺎﺿﺮ‬.‫ ﻛﻤﺎ ﺃﻥ ﻟﺪﻳﻬﻢ ﺃﻳﻀًﺎ ﺍﻫﺘﻤﺎﻣﺎﺕ ﺿﻴﻘﺔ ﻭﺇﺟﺮﺍﺀﺍﺕ ﺭﻭﺗﻴﻨﻴﺔ‬،‫ﺻﻌﻮﺑﺔ ﻓﻲ ﺍﻟﺘﻔﺎﻋﻞ ﻣﻊ ﺍﻵﺧﺮﻳﻦ ﻭﺍﻟﺘﻮﺍﺻﻞ‬
[29].‫ﺗﺤﺪﻳﺪ ﺍﺿﻄﺮﺍﺏ ﻃﻴﻒ ﺍﻟﺘﻮﺣﺪ ﻣﺸﻜﻠﺔ ﻛﺒﻴﺮﺓ ﻓﻲ ﻣﺠﺎﻝ ﺍﻛﺘﺸﺎﻑ ﺍﻷﻣﺮﺍﺽ ﺍﻟﻤﺘﻘﺪﻣﺔ‬

The notion of identification pertains to a methodology that relies on contemporary data as input
to generate estimations that discern forthcoming patterns, with the aim of recognising, detecting,
or preempting occurrences. The identification system refers to a safety mechanism that records
changes in data or statistics, facilitating the analysis of interrelationships among various
variables and enabling informed decision-making [30].

‫ ﺑﻬﺪﻑ ﺍﻟﺘﻌﺮﻑ‬،‫ﺗﺘﻌﻠﻖ ﻓﻜﺮﺓ ﺗﺤﺪﻳﺪ ﺍﻟﻬﻮﻳﺔ ﺑﻤﻨﻬﺠﻴﺔ ﺗﻌﺘﻤﺪ ﻋﻠﻰ ﺍﻟﺒﻴﺎﻧﺎﺕ ﺍﻟﻤﻌﺎﺻﺮﺓ ﻛﻤﺪﺧﻼﺕ ﻟﺘﻮﻟﻴﺪ ﺗﻘﺪﻳﺮﺍﺕ ﺗﻤﻴﺰ ﺍﻷﻧﻤﺎﻁ ﺍﻟﻘﺎﺩﻣﺔ‬
‫ ﻣﻤﺎ‬،‫ ﻳﺸﻴﺮ ﻧﻈﺎﻡ ﺗﺤﺪﻳﺪ ﺍﻟﻬﻮﻳﺔ ﺇﻟﻰ ﺁﻟﻴﺔ ﺃﻣﺎﻥ ﺗﺴﺠﻞ ﺍﻟﺘﻐﻴﺮﺍﺕ ﻓﻲ ﺍﻟﺒﻴﺎﻧﺎﺕ ﺃﻭ ﺍﻹﺣﺼﺎﺋﻴﺎﺕ‬.‫ﻋﻠﻰ ﺍﻷﺣﺪﺍﺙ ﺃﻭ ﺍﻛﺘﺸﺎﻓﻬﺎ ﺃﻭ ﺍﺳﺘﺒﺎﻗﻬﺎ‬
.[30] ‫ﻳﺴﻬﻞ ﺗﺤﻠﻴﻞ ﺍﻟﻌﻼﻗﺎﺕ ﺍﻟﻤﺘﺒﺎﺩﻟﺔ ﺑﻴﻦ ﺍﻟﻤﺘﻐﻴﺮﺍﺕ ﺍﻟﻤﺨﺘﻠﻔﺔ ﻭﻳﺘﻴﺢ ﺍﺗﺨﺎﺫ ﻗﺮﺍﺭﺍﺕ ﻣﺴﺘﻨﻴﺮﺓ‬

There are many solutions for disease identification, including machine learning. Machine
learning is a new field of research that deals with finding theories and algorithms that allow the
machine to learn on its own. Machine learning, being a subset of artificial intelligence, mostly
focuses its research efforts on the identification of robust methodologies via the examination of
extensive datasets. This chapter provides an overview of the idea of autism spectrum disorder. The
topic of discussion includes the concept of preprocessing, together with the machine learning
component and the techniques used. This chapter elucidates the ASD identification system as a
framework predicated on the assessment of ASD risk and the subsequent detection of potential
cases, with the ultimate goal of implementing measures to mitigate the impact of ASD.
1
‫ ﻳﻌﺪ ﺍﻟﺘﻌﻠﻢ ﺍﻵﻟﻲ ﻣﺠﺎﻟًﺎ ﺟﺪﻳﺪًﺍ ﻟﻠﺒﺤﺚ ﻳﺘﻌﺎﻣﻞ ﻣﻊ ﺇﻳﺠﺎﺩ ﺍﻟﻨﻈﺮﻳﺎﺕ‬.‫ ﺑﻤﺎ ﻓﻲ ﺫﻟﻚ ﺍﻟﺘﻌﻠﻢ ﺍﻵﻟﻲ‬،‫ﻫﻨﺎﻙ ﺍﻟﻌﺪﻳﺪ ﻣﻦ ﺍﻟﺤﻠﻮﻝ ﻟﺘﺤﺪﻳﺪ ﺍﻷﻣﺮﺍﺽ‬
‫ ﻳﺮﻛﺰ ﻓﻲ‬،‫ ﻛﻮﻧﻪ ﻣﺠﻤﻮﻋﺔ ﻓﺮﻋﻴﺔ ﻣﻦ ﺍﻟﺬﻛﺎﺀ ﺍﻻﺻﻄﻨﺎﻋﻲ‬،‫ ﺍﻟﺘﻌﻠﻢ ﺍﻵﻟﻲ‬.‫ﻭﺍﻟﺨﻮﺍﺭﺯﻣﻴﺎﺕ ﺍﻟﺘﻲ ﺗﺴﻤﺢ ﻟﻶﻟﺔ ﺑﺎﻟﺘﻌﻠﻢ ﻣﻦ ﺗﻠﻘﺎﺀ ﻧﻔﺴﻬﺎ‬
‫ ﻳﻘﺪﻡ ﻫﺬﺍ ﺍﻟﻔﺼﻞ ﻟﻤﺤﺔ ﻋﺎﻣﺔ ﻋﻦ ﻓﻜﺮﺓ‬.‫ﺍﻟﻐﺎﻟﺐ ﺟﻬﻮﺩﻩ ﺍﻟﺒﺤﺜﻴﺔ ﻋﻠﻰ ﺗﺤﺪﻳﺪ ﻣﻨﻬﺠﻴﺎﺕ ﻗﻮﻳﺔ ﻣﻦ ﺧﻼﻝ ﻓﺤﺺ ﻣﺠﻤﻮﻋﺎﺕ ﺍﻟﺒﻴﺎﻧﺎﺕ ﺍﻟﺸﺎﻣﻠﺔ‬
.‫ ﺑﺎﻹﺿﺎﻓﺔ ﺇﻟﻰ ﻣﻜﻮﻥ ﺍﻟﺘﻌﻠﻢ ﺍﻵﻟﻲ ﻭﺍﻟﺘﻘﻨﻴﺎﺕ ﺍﻟﻤﺴﺘﺨﺪﻣﺔ‬،‫ ﻳﺘﻀﻤﻦ ﻣﻮﺿﻮﻉ ﺍﻟﻤﻨﺎﻗﺸﺔ ﻣﻔﻬﻮﻡ ﺍﻟﻤﻌﺎﻟﺠﺔ ﺍﻟﻤﺴﺒﻘﺔ‬.‫ﺍﺿﻄﺮﺍﺏ ﻃﻴﻒ ﺍﻟﺘﻮﺣﺪ‬
‫ﻳﻮﺿﺢ ﻫﺬﺍ ﺍﻟﻔﺼﻞ ﻧﻈﺎﻡ ﺗﺤﺪﻳﺪ ﺍﺿﻄﺮﺍﺏ ﻃﻴﻒ ﺍﻟﺘﻮﺣﺪ ﻛﺈﻃﺎﺭ ﻳﻘﻮﻡ ﻋﻠﻰ ﺗﻘﻴﻴﻢ ﻣﺨﺎﻃﺮ ﺍﺿﻄﺮﺍﺏ ﻃﻴﻒ ﺍﻟﺘﻮﺣﺪ ﻭﺍﻟﻜﺸﻒ ﺍﻟﻼﺣﻖ ﻋﻦ‬
.‫ ﻣﻊ ﺍﻟﻬﺪﻑ ﺍﻟﻨﻬﺎﺋﻲ ﺍﻟﻤﺘﻤﺜﻞ ﻓﻲ ﺗﻨﻔﻴﺬ ﺗﺪﺍﺑﻴﺮ ﻟﻠﺘﺨﻔﻴﻒ ﻣﻦ ﺗﺄﺛﻴﺮ ﺍﺿﻄﺮﺍﺏ ﻃﻴﻒ ﺍﻟﺘﻮﺣﺪ‬،‫ﺍﻟﺤﺎﻻﺕ ﺍﻟﻤﺤﺘﻤﻠﺔ‬

2.2 autism spectrum disorder

Autism Spectrum Disorder, a dangerous condition that may impede one's capacity to connect or
communicate with others, develops throughout a person's formative years. It affects the
neurological system and is often brought on by heredity or environmental circumstances. As a
consequence, it has an influence on an individual's entire cognitive, social, emotional, and
physical health[31].

‫ ﺍﺿﻄﺮﺍﺏ ﻃﻴﻒ ﺍﻟﺘﻮﺣﺪ‬2.2

‫ ﻳﺘﻄﻮﺭ ﻃﻮﺍﻝ ﺳﻨﻮﺍﺕ‬،‫ ﻭﻫﻮ ﺣﺎﻟﺔ ﺧﻄﻴﺮﺓ ﻗﺪ ﺗﻌﻴﻖ ﻗﺪﺭﺓ ﺍﻟﺸﺨﺺ ﻋﻠﻰ ﺍﻻﺗﺼﺎﻝ ﺃﻭ ﺍﻟﺘﻮﺍﺻﻞ ﻣﻊ ﺍﻵﺧﺮﻳﻦ‬،‫ﺍﺿﻄﺮﺍﺏ ﻃﻴﻒ ﺍﻟﺘﻮﺣﺪ‬
‫ ﻓﺈﻥ ﻟﻪ ﺗﺄﺛﻴﺮًﺍ‬،‫ ﻭﻧﺘﻴﺠﺔ ﻟﺬﻟﻚ‬.‫ ﻭﻫﻮ ﻳﺆﺛﺮ ﻋﻠﻰ ﺍﻟﺠﻬﺎﺯ ﺍﻟﻌﺼﺒﻲ ﻭﻏﺎﻟﺒﺎً ﻣﺎ ﻳﺤﺪﺙ ﺑﺴﺒﺐ ﺍﻟﻮﺭﺍﺛﺔ ﺃﻭ ﺍﻟﻈﺮﻭﻑ ﺍﻟﺒﻴﺌﻴﺔ‬.‫ﺗﻜﻮﻳﻦ ﺍﻟﺸﺨﺺ‬
[31].‫ﻋﻠﻰ ﺻﺤﺔ ﺍﻟﻔﺮﺩ ﺍﻟﻤﻌﺮﻓﻴﺔ ﻭﺍﻻﺟﺘﻤﺎﻋﻴﺔ ﻭﺍﻟﻌﺎﻃﻔﻴﺔ ﻭﺍﻟﺠﺴﺪﻳﺔ ﺑﺄﻛﻤﻠﻬﺎ‬

The breadth of its symptoms, as well as how severe they are, vary greatly. A few of the typical
symptoms the person experiences include communication problems, particularly in social
situations, obsessional hobbies, and repeated mannerisms. A thorough examination is necessary to
detect ASD. This also comprises a thorough examination as well as numerous tests conducted by
child psychologists and other qualified experts. Autism Diagnostic Interview Revised (ADI-R)
and Autism Diagnostic Observation Schedule Revised (ADOS-R) are examples of traditional
diagnostic techniques. They, however, take a longer time and effort since they are long and
difficult. The children population is significantly affected by ASD. In the majority of instances, it
can typically be detected in its early phases, but the main bottleneck is caused by the subjective
and time-consuming nature of current diagnostic approaches. [32]

2
 ‫ﻭﻳﺨﺘﻠﻒ ﺍﺗﺴﺎﻉ ﻧﻄﺎﻕ ﺃﻋﺮﺍﺿﻪ‪ ،‬ﻭﻛﺬﻟﻚ ﻣﺪﻯ ﺷﺪﺗﻬﺎ‪ ،‬ﺑﺸﻜﻞ ﻛﺒﻴﺮ‪ .‬ﺗﺸﻤﻞ ﺑﻌﺾ ﺍﻷﻋﺮﺍﺽ ﺍﻟﻨﻤﻮﺫﺟﻴﺔ ﺍﻟﺘﻲ ﻳﻌﺎﻧﻲ ﻣﻨﻬﺎ ﺍﻟﺸﺨﺺ‬
‫ﻣﺸﺎﻛﻞ ﺍﻟﺘﻮﺍﺻﻞ‪ ،‬ﺧﺎﺻﺔ ﻓﻲ ﺍﻟﻤﻮﺍﻗﻒ ﺍﻻﺟﺘﻤﺎﻋﻴﺔ‪ ،‬ﻭﺍﻟﻬﻮﺍﻳﺎﺕ ﺍﻟﻮﺳﻮﺍﺳﻴﺔ‪ ،‬ﻭﺍﻟﺴﻠﻮﻛﻴﺎﺕ ﺍﻟﻤﺘﻜﺮﺭﺓ‪ .‬ﻣﻦ ﺍﻟﻀﺮﻭﺭﻱ ﺇﺟﺮﺍﺀ ﻓﺤﺺ‬
‫ﺷﺎﻣﻞ ﻟﻠﻜﺸﻒ ﻋﻦ ﺍﺿﻄﺮﺍﺏ ﻃﻴﻒ ﺍﻟﺘﻮﺣﺪ‪ .‬ﻭﻳﺘﻀﻤﻦ ﻫﺬﺍ ﺃﻳﻀًﺎ ﻓﺤﺼًﺎ ﺷﺎﻣﻼً ﺑﺎﻹﺿﺎﻓﺔ ﺇﻟﻰ ﺍﻟﻌﺪﻳﺪ ﻣﻦ ﺍﻻﺧﺘﺒﺎﺭﺍﺕ ﺍﻟﺘﻲ ﻳﺠﺮﻳﻬﺎ‬
‫ﻭﺟﺪﻭﻝ ﻣﺮﺍﻗﺒﺔ ﺗﺸﺨﻴﺺ )‪ (ADI-R‬ﻋﻠﻤﺎﺀ ﻧﻔﺲ ﺍﻷﻃﻔﺎﻝ ﻭﻏﻴﺮﻫﻢ ﻣﻦ ﺍﻟﺨﺒﺮﺍﺀ ﺍﻟﻤﺆﻫﻠﻴﻦ‪ .‬ﺗﻌﺘﺒﺮ ﻣﻘﺎﺑﻠﺔ ﺗﺸﺨﻴﺺ ﺍﻟﺘﻮﺣﺪ ﺍﻟﻤﻨﻘﺤﺔ‬
‫ﺃﻣﺜﻠﺔ ﻋﻠﻰ ﺗﻘﻨﻴﺎﺕ ﺍﻟﺘﺸﺨﻴﺺ ﺍﻟﺘﻘﻠﻴﺪﻳﺔ‪ .‬ﻭﻣﻊ ﺫﻟﻚ‪ ،‬ﻓﻬﻲ ﺗﺴﺘﻐﺮﻕ ﻭﻗﺘًﺎ ﻭﺟﻬﺪًﺍ ﺃﻃﻮﻝ ﻷﻧﻬﺎ ﻃﻮﻳﻠﺔ )‪ (ADOS-R‬ﺍﻟﺘﻮﺣﺪ ﺍﻟﻤﻨﻘﺢ‬
‫ﻭﺻﻌﺒﺔ‪ .‬ﻳﺘﺄﺛﺮ ﻋﺪﺩ ﺍﻷﻃﻔﺎﻝ ﺑﺸﻜﻞ ﻛﺒﻴﺮ ﺑﺎﺿﻄﺮﺍﺏ ﻃﻴﻒ ﺍﻟﺘﻮﺣﺪ‪ .‬ﻓﻲ ﻣﻌﻈﻢ ﺍﻟﺤﺎﻻﺕ‪ ،‬ﻳﻤﻜﻦ ﻋﺎﺩﺓً ﺍﻛﺘﺸﺎﻓﻪ ﻓﻲ ﻣﺮﺍﺣﻠﻪ ﺍﻟﻤﺒﻜﺮﺓ‪،‬‬
‫ﺍﻟﺮﺋﻴﺴﻲ ﻫﻮ ﺍﻟﻄﺒﻴﻌﺔ ﺍﻟﺬﺍﺗﻴﺔ ﻭﺍﻟﻤﺴﺘﻬﻠﻜﺔ ﻟﻠﻮﻗﺖ ﻷﺳﺎﻟﻴﺐ ﺍﻟﺘﺸﺨﻴﺺ ﺍﻟﺤﺎﻟﻴﺔ]‪. [32‬ﻭﻟﻜﻦ ﺳﺒﺐ ﺍﻻﺧﺘﻨﺎﻕ‬

‫‪Treatment of autism spectrum disorder symptoms is greatly aided by early diagnosis. Artificial‬‬
‫‪intelligence and machine learning have made great strides in recent years. Research in these areas‬‬
‫‪has helped advance health care considerably. Research on early autism diagnosis has received a‬‬
‫‪lot of attention. Researchers are applying a variety of machine learning algorithms to identify‬‬
‫‪autistic traits in both younger and older populations. In recent years, image classification and‬‬
‫‪magnetic resonance imaging (MRI) pictures have been employed with deep learning approaches‬‬
‫‪to identify autism. To further improve autism identification using machine learning and deep‬‬
‫]‪learning methods, several datasets have been produced.[33‬‬

‫ﺇﻥ ﻋﻼﺝ ﺃﻋﺮﺍﺽ ﺍﺿﻄﺮﺍﺏ ﻃﻴﻒ ﺍﻟﺘﻮﺣﺪ ﻳﺴﺎﻋﺪ ﺑﺸﻜﻞ ﻛﺒﻴﺮ ﻓﻲ ﺍﻟﺘﺸﺨﻴﺺ ﺍﻟﻤﺒﻜﺮ‪ .‬ﺣﻘﻖ ﺍﻟﺬﻛﺎﺀ ﺍﻻﺻﻄﻨﺎﻋﻲ ﻭﺍﻟﺘﻌﻠﻢ ﺍﻵﻟﻲ‬
‫ﺧﻄﻮﺍﺕ ﻛﺒﻴﺮﺓ ﻓﻲ ﺍﻟﺴﻨﻮﺍﺕ ﺍﻷﺧﻴﺮﺓ‪ .‬ﻭﻗﺪ ﺳﺎﻋﺪﺕ ﺍﻷﺑﺤﺎﺙ ﻓﻲ ﻫﺬﻩ ﺍﻟﻤﺠﺎﻻﺕ ﻋﻠﻰ ﺗﻄﻮﻳﺮ ﺍﻟﺮﻋﺎﻳﺔ ﺍﻟﺼﺤﻴﺔ ﺑﺸﻜﻞ ﻛﺒﻴﺮ‪ .‬ﺣﻈﻴﺖ‬
‫ﺍﻷﺑﺤﺎﺙ ﺍﻟﻤﺘﻌﻠﻘﺔ ﺑﺎﻟﺘﺸﺨﻴﺺ ﺍﻟﻤﺒﻜﺮ ﻟﻤﺮﺽ ﺍﻟﺘﻮﺣﺪ ﺑﺎﻟﻜﺜﻴﺮ ﻣﻦ ﺍﻻﻫﺘﻤﺎﻡ‪ .‬ﻳﻄﺒﻖ ﺍﻟﺒﺎﺣﺜﻮﻥ ﻣﺠﻤﻮﻋﺔ ﻣﺘﻨﻮﻋﺔ ﻣﻦ ﺧﻮﺍﺭﺯﻣﻴﺎﺕ ﺍﻟﺘﻌﻠﻢ‬
‫ﺍﻵﻟﻲ ﻟﺘﺤﺪﻳﺪ ﺳﻤﺎﺕ ﺍﻟﺘﻮﺣﺪ ﻟﺪﻯ ﻛﻞ ﻣﻦ ﺍﻟﺴﻜﺎﻥ ﺍﻷﺻﻐﺮ ﺳﻨﺎ ﻭﻛﺒﺎﺭ ﺍﻟﺴﻦ‪ .‬ﻓﻲ ﺍﻟﺴﻨﻮﺍﺕ ﺍﻷﺧﻴﺮﺓ‪ ،‬ﺗﻢ ﺍﺳﺘﺨﺪﺍﻡ ﺗﺼﻨﻴﻒ ﺍﻟﺼﻮﺭ‬
‫ﻣﻊ ﺃﺳﺎﻟﻴﺐ ﺍﻟﺘﻌﻠﻢ ﺍﻟﻌﻤﻴﻖ ﻟﺘﺤﺪﻳﺪ ﻣﺮﺽ ﺍﻟﺘﻮﺣﺪ‪ .‬ﻟﻤﺰﻳﺪ ﻣﻦ ﺗﺤﺴﻴﻦ ﺍﻟﺘﻌﺮﻑ ﻋﻠﻰ )‪ (MRI‬ﻭﺻﻮﺭ ﺍﻟﺘﺼﻮﻳﺮ ﺑﺎﻟﺮﻧﻴﻦ ﺍﻟﻤﻐﻨﺎﻃﻴﺴﻲ‬
‫ﺍﻟﺘﻮﺣﺪ ﺑﺎﺳﺘﺨﺪﺍﻡ ﺍﻟﺘﻌﻠﻢ ﺍﻵﻟﻲ ﻭﺃﺳﺎﻟﻴﺐ ﺍﻟﺘﻌﻠﻢ ﺍﻟﻌﻤﻴﻖ‪ ،‬ﺗﻢ ﺇﻧﺘﺎﺝ ﺍﻟﻌﺪﻳﺪ ﻣﻦ ﻣﺠﻤﻮﻋﺎﺕ ﺍﻟﺒﻴﺎﻧﺎﺕ]‪[33‬‬

‫‪2.3 Type of Autism‬‬

‫‪3‬‬
Classifying autism in children can be complex due to the different diagnostic systems practitioners
use. The most commonly recognized subtypes of autism are:

2.3 ‫ﻧﻮﻉ ﺍﻟﺘﻮﺣﺪ‬

‫ ﺍﻷﻧﻮﺍﻉ‬.‫ﻳﻤﻜﻦ ﺃﻥ ﻳﻜﻮﻥ ﺗﺼﻨﻴﻒ ﻣﺮﺽ ﺍﻟﺘﻮﺣﺪ ﻟﺪﻯ ﺍﻷﻃﻔﺎﻝ ﻣﻌﻘﺪًﺍ ﺑﺴﺒﺐ ﺍﺧﺘﻼﻑ ﺃﻧﻈﻤﺔ ﺍﻟﺘﺸﺨﻴﺺ ﺍﻟﺘﻲ ﻳﺴﺘﺨﺪﻣﻬﺎ ﺍﻟﻤﻤﺎﺭﺳﻮﻥ‬
‫ﺍﻟﻔﺮﻋﻴﺔ ﺍﻷﻛﺜﺮ ﺷﻴﻮﻋًﺎ ﻟﻤﺮﺽ ﺍﻟﺘﻮﺣﺪ ﻫﻲ‬
2.3.1 Autistic spectrum disorder
Autism spectrum disorder is a term used to describe a constellation of early-appearing social
communication deficits and repetitive sensory–motor behavior’s associated with a strong genetic
component as well as other causes. The outlook for many individuals with autism spectrum
disorder today is brighter than it was 50 years ago[34].

‫ ﺍﺿﻄﺮﺍﺏ ﻃﻴﻒ ﺍﻟﺘﻮﺣﺪ‬2.3.1

‫ﺍﺿﻄﺮﺍﺏ ﻃﻴﻒ ﺍﻟﺘﻮﺣﺪ ﻫﻮ ﻣﺼﻄﻠﺢ ﻳﺴﺘﺨﺪﻡ ﻟﻮﺻﻒ ﻣﺠﻤﻮﻋﺔ ﻣﻦ ﻋﺠﺰ ﺍﻟﺘﻮﺍﺻﻞ ﺍﻻﺟﺘﻤﺎﻋﻲ ﺍﻟﺬﻱ ﻳﻈﻬﺮ ﻣﺒﻜﺮًﺍ ﻭﺍﻟﺴﻠﻮﻙ ﺍﻟﺤﺴﻲ‬
‫ ﺇﻥ ﺍﻟﺘﻮﻗﻌﺎﺕ ﺑﺎﻟﻨﺴﺒﺔ ﻟﻠﻌﺪﻳﺪ ﻣﻦ ﺍﻷﻓﺮﺍﺩ ﺍﻟﻤﺼﺎﺑﻴﻦ‬.‫ﺍﻟﺤﺮﻛﻲ ﺍﻟﻤﺘﻜﺮﺭ ﺍﻟﻤﺮﺗﺒﻂ ﺑﻤﻜﻮﻥ ﻭﺭﺍﺛﻲ ﻗﻮﻱ ﺑﺎﻹﺿﺎﻓﺔ ﺇﻟﻰ ﺃﺳﺒﺎﺏ ﺃﺧﺮﻯ‬
.[34]‫ ﻋﺎﻣًﺎ‬50 ‫ﺑﺎﺿﻄﺮﺍﺏ ﻃﻴﻒ ﺍﻟﺘﻮﺣﺪ ﺍﻟﻴﻮﻡ ﺃﻛﺜﺮ ﺇﺷﺮﺍﻗًﺎ ﻣﻤﺎ ﻛﺎﻧﺖ ﻋﻠﻴﻪ ﻗﺒﻞ‬
Over the course of the last five decades, there has been a notable transformation in the
understanding of autism spectrum disorder. Initially characterised as an uncommon childhood
illness with certain parameters. ASD has now garnered significant attention, advocacy, and study,
leading to the recognition of its prevalence and considerable heterogeneity across an individual's
lifespan. The essential symptoms of Autism Spectrum Disorderhave been rather consistent since
its first characterization, including impairments in social communication and atypical sensory-
motor behaviours. Nevertheless, contemporary understanding of autism acknowledges it as a
spectrum with a broad variety of manifestations, varying in severity from mild to severe.
However, it is important to note that a significant proportion of persons diagnosed with Autism
Spectrum Disorder require ongoing assistance throughout their lives[35].

‫ ﺗﻢ ﻭﺻﻔﻪ ﻓﻲ ﺍﻟﺒﺪﺍﻳﺔ ﺑﺄﻧﻪ ﻣﺮﺽ ﻏﻴﺮ‬.‫ ﺣﺪﺙ ﺗﺤﻮﻝ ﻣﻠﺤﻮﻅ ﻓﻲ ﻓﻬﻢ ﺍﺿﻄﺮﺍﺏ ﻃﻴﻒ ﺍﻟﺘﻮﺣﺪ‬،‫ﻋﻠﻰ ﻣﺪﺍﺭ ﺍﻟﻌﻘﻮﺩ ﺍﻟﺨﻤﺴﺔ ﺍﻟﻤﺎﺿﻴﺔ‬
‫ ﻣﻤﺎ ﺃﺩﻯ ﺇﻟﻰ‬،‫ ﻟﻘﺪ ﺣﻈﻲ ﺍﺿﻄﺮﺍﺏ ﻃﻴﻒ ﺍﻟﺘﻮﺣﺪ ﺍﻵﻥ ﺑﺎﻫﺘﻤﺎﻡ ﻛﺒﻴﺮ ﻭﻣﻨﺎﺻﺮﺓ ﻭﺩﺭﺍﺳﺔ‬.‫ﺷﺎﺋﻊ ﻓﻲ ﻣﺮﺣﻠﺔ ﺍﻟﻄﻔﻮﻟﺔ ﻣﻊ ﻣﻌﺎﻳﻴﺮ ﻣﻌﻴﻨﺔ‬
‫ ﻛﺎﻧﺖ ﺍﻷﻋﺮﺍﺽ ﺍﻷﺳﺎﺳﻴﺔ ﻻﺿﻄﺮﺍﺏ ﻃﻴﻒ ﺍﻟﺘﻮﺣﺪ ﻣﺘﺴﻘﺔ ﺇﻟﻰ ﺣﺪ ﻣﺎ‬.‫ﺍﻻﻋﺘﺮﺍﻑ ﺑﺎﻧﺘﺸﺎﺭﻩ ﻭﻋﺪﻡ ﺗﺠﺎﻧﺴﻪ ﺍﻟﻜﺒﻴﺮ ﻋﺒﺮ ﻋﻤﺮ ﺍﻟﻔﺮﺩ‬
‫ ﻓﺈﻥ‬،‫ ﻭﻣﻊ ﺫﻟﻚ‬.‫ ﺑﻤﺎ ﻓﻲ ﺫﻟﻚ ﺿﻌﻒ ﻓﻲ ﺍﻟﺘﻮﺍﺻﻞ ﺍﻻﺟﺘﻤﺎﻋﻲ ﻭﺍﻟﺴﻠﻮﻛﻴﺎﺕ ﺍﻟﺤﺴﻴﺔ ﺍﻟﺤﺮﻛﻴﺔ ﻏﻴﺮ ﺍﻟﻨﻤﻄﻴﺔ‬،‫ﻣﻨﺬ ﺗﺸﺨﻴﺼﻪ ﻷﻭﻝ ﻣﺮﺓ‬
4
‫ ﺗﺘﻔﺎﻭﺕ ﻓﻲ ﺷﺪﺗﻬﺎ ﻣﻦ ﺧﻔﻴﻔﺔ‬،‫ﺍﻟﻔﻬﻢ ﺍﻟﻤﻌﺎﺻﺮ ﻟﻤﺮﺽ ﺍﻟﺘﻮﺣﺪ ﻳﻌﺘﺮﻑ ﺑﻪ ﺑﺎﻋﺘﺒﺎﺭﻩ ﻃﻴﻔًﺎ ﻳﺸﺘﻤﻞ ﻋﻠﻰ ﻣﺠﻤﻮﻋﺔ ﻭﺍﺳﻌﺔ ﻣﻦ ﺍﻟﻤﻈﺎﻫﺮ‬
‫ ﻓﻤﻦ ﺍﻟﻤﻬﻢ ﻣﻼﺣﻈﺔ ﺃﻥ ﻧﺴﺒﺔ ﻛﺒﻴﺮﺓ ﻣﻦ ﺍﻷﺷﺨﺎﺹ ﺍﻟﺬﻳﻦ ﺗﻢ ﺗﺸﺨﻴﺺ ﺇﺻﺎﺑﺘﻬﻢ ﺑﺎﺿﻄﺮﺍﺏ ﻃﻴﻒ ﺍﻟﺘﻮﺣﺪ‬،‫ ﻭﻣﻊ ﺫﻟﻚ‬.‫ﺇﻟﻰ ﺷﺪﻳﺪﺓ‬
.[35]‫ﻳﺤﺘﺎﺟﻮﻥ ﺇﻟﻰ ﻣﺴﺎﻋﺪﺓ ﻣﺴﺘﻤﺮﺓ ﻃﻮﺍﻝ ﺣﻴﺎﺗﻬﻢ‬
The fields of genetics and neurology have successfully found patterns of risk that are very
fascinating. However, it is important to note that these patterns have not yet yielded significant
practical benefits. Further research is required to get a comprehensive understanding of the
optimal conditions in which behavioural and medical interventions might yield positive outcomes.
This includes determining the specific population of children, particularly those with significant
comorbidities, for whom these therapies are most beneficial. Clinicians possess the capacity to
effect positive change through the provision of prompt and personalised assistance to families as
they navigate the process of obtaining referrals and accessing community support systems.
Additionally, clinicians play a crucial role in disseminating accurate information, despite the
potential influence of unfiltered media sources. Furthermore, clinicians are able to anticipate and
address various transitions, such as changes within the family dynamic and the entry or departure
of a child from the school environment [34].

‫ ﻓﻤﻦ ﺍﻟﻤﻬﻢ ﺃﻥ ﻧﻼﺣﻆ ﺃﻥ‬،‫ ﻭﻣﻊ ﺫﻟﻚ‬.‫ﻟﻘﺪ ﻧﺠﺤﺖ ﻣﺠﺎﻻﺕ ﻋﻠﻢ ﺍﻟﻮﺭﺍﺛﺔ ﻭﻋﻠﻢ ﺍﻷﻋﺼﺎﺏ ﻓﻲ ﺍﻟﻌﺜﻮﺭ ﻋﻠﻰ ﺃﻧﻤﺎﻁ ﺧﻄﺮ ﺭﺍﺋﻌﺔ ﻟﻠﻐﺎﻳﺔ‬
‫ ﻫﻨﺎﻙ ﺣﺎﺟﺔ ﺇﻟﻰ ﻣﺰﻳﺪ ﻣﻦ ﺍﻟﺒﺤﺚ ﻟﻠﺤﺼﻮﻝ ﻋﻠﻰ ﻓﻬﻢ ﺷﺎﻣﻞ ﻟﻠﻈﺮﻭﻑ ﺍﻟﻤﺜﺎﻟﻴﺔ‬.‫ﻫﺬﻩ ﺍﻷﻧﻤﺎﻁ ﻟﻢ ﺗﺴﻔﺮ ﺑﻌﺪ ﻋﻦ ﻓﻮﺍﺋﺪ ﻋﻤﻠﻴﺔ ﻛﺒﻴﺮﺓ‬
‫ ﻭﺧﺎﺻﺔ‬،‫ ﻳﺘﻀﻤﻦ ﺫﻟﻚ ﺗﺤﺪﻳﺪ ﻣﺠﻤﻮﻋﺔ ﻣﺤﺪﺩﺓ ﻣﻦ ﺍﻷﻃﻔﺎﻝ‬.‫ﺍﻟﺘﻲ ﻗﺪ ﺗﺆﺩﻱ ﻓﻴﻬﺎ ﺍﻟﺘﺪﺧﻼﺕ ﺍﻟﺴﻠﻮﻛﻴﺔ ﻭﺍﻟﻄﺒﻴﺔ ﺇﻟﻰ ﻧﺘﺎﺋﺞ ﺇﻳﺠﺎﺑﻴﺔ‬
‫ ﻳﻤﺘﻠﻚ ﺍﻷﻃﺒﺎﺀ ﺍﻟﻘﺪﺭﺓ ﻋﻠﻰ ﺇﺣﺪﺍﺙ‬.‫ ﻭﺍﻟﺬﻳﻦ ﺗﻜﻮﻥ ﻫﺬﻩ ﺍﻟﻌﻼﺟﺎﺕ ﻣﻔﻴﺪﺓ ﻟﻬﻢ ﺃﻛﺜﺮ‬،‫ﺃﻭﻟﺌﻚ ﺍﻟﺬﻳﻦ ﻳﻌﺎﻧﻮﻥ ﻣﻦ ﺃﻣﺮﺍﺽ ﻣﺼﺎﺣﺒﺔ ﻛﺒﻴﺮﺓ‬
‫ﺗﻐﻴﻴﺮ ﺇﻳﺠﺎﺑﻲ ﻣﻦ ﺧﻼﻝ ﺗﻮﻓﻴﺮ ﺍﻟﻤﺴﺎﻋﺪﺓ ﺍﻟﺴﺮﻳﻌﺔ ﻭﺍﻟﺸﺨﺼﻴﺔ ﻟﻠﻌﺎﺋﻼﺕ ﺃﺛﻨﺎﺀ ﺗﻨﻘﻠﻬﻢ ﻓﻲ ﻋﻤﻠﻴﺔ ﺍﻟﺤﺼﻮﻝ ﻋﻠﻰ ﺍﻹﺣﺎﻻﺕ ﻭﺍﻟﻮﺻﻮﻝ‬
‫ ﻋﻠﻰ ﺍﻟﺮﻏﻢ ﻣﻦ ﺍﻟﺘﺄﺛﻴﺮ ﺍﻟﻤﺤﺘﻤﻞ‬،‫ ﻳﻠﻌﺐ ﺍﻷﻃﺒﺎﺀ ﺩﻭﺭًﺍ ﺣﺎﺳﻤًﺎ ﻓﻲ ﻧﺸﺮ ﻣﻌﻠﻮﻣﺎﺕ ﺩﻗﻴﻘﺔ‬،‫ ﺑﺎﻹﺿﺎﻓﺔ ﺇﻟﻰ ﺫﻟﻚ‬.‫ﺇﻟﻰ ﺃﻧﻈﻤﺔ ﺩﻋﻢ ﺍﻟﻤﺠﺘﻤﻊ‬
‫ ﻣﺜﻞ ﺍﻟﺘﻐﻴﺮﺍﺕ ﺩﺍﺧﻞ‬،‫ ﻳﺴﺘﻄﻴﻊ ﺍﻷﻃﺒﺎﺀ ﺗﻮﻗﻊ ﻭﻣﻌﺎﻟﺠﺔ ﺍﻟﺘﺤﻮﻻﺕ ﺍﻟﻤﺨﺘﻠﻔﺔ‬،‫ ﻋﻼﻭﺓ ﻋﻠﻰ ﺫﻟﻚ‬.‫ﻟﻤﺼﺎﺩﺭ ﺍﻟﻮﺳﺎﺋﻂ ﻏﻴﺮ ﺍﻟﻤﺼﻔﺎﺓ‬
[34] ‫ﺩﻳﻨﺎﻣﻴﻜﻴﺔ ﺍﻷﺳﺮﺓ ﻭﺩﺧﻮﻝ ﺍﻟﻄﻔﻞ ﺃﻭ ﺧﺮﻭﺟﻪ ﻣﻦ ﺍﻟﺒﻴﺌﺔ ﺍﻟﻤﺪﺭﺳﻴﺔ‬
2.3.2 Pathological Demand Avoidance
The term "Pathological Demand Avoidance" (PDA) was used during the 1980s to characterise
children who fall within the spectrum of pervasive developmental disorders. These children
exhibit a compulsive aversion to routine demands, an intense need for control, and a seemingly
limited capacity for social identification, self-esteem, or feelings of guilt. The term PDA has
garnered significant attention and generated substantial debate in academic circles.[36]
5
 ‫ ﺗﺠﻨﺐ ﺍﻟﻄﻠﺐ ﺍﻟﻤﺮﺿﻲ‬2.3.2
‫( ﺧﻼﻝ ﺍﻟﺜﻤﺎﻧﻴﻨﻴﺎﺕ ﻟﻮﺻﻒ ﺍﻷﻃﻔﺎﻝ ﺍﻟﺬﻳﻦ ﻳﻘﻌﻮﻥ ﺿﻤﻦ ﻧﻄﺎﻕ ﺍﺿﻄﺮﺍﺑﺎﺕ‬PDA) "‫ﺗﻢ ﺍﺳﺘﺨﺪﺍﻡ ﻣﺼﻄﻠﺢ "ﺗﺠﻨﺐ ﺍﻟﻄﻠﺐ ﺍﻟﻤﺮﺿﻲ‬
‫ ﻭﻗﺪﺭﺓ ﻣﺤﺪﻭﺩﺓ ﻋﻠﻰ ﻣﺎ ﻳﺒﺪﻭ‬،‫ ﻭﺣﺎﺟﺔ ﺷﺪﻳﺪﺓ ﻟﻠﺴﻴﻄﺮﺓ‬،‫ ﻳﻈﻬﺮ ﻫﺆﻻﺀ ﺍﻷﻃﻔﺎﻝ ﻧﻔﻮﺭًﺍ ﻗﻬﺮﻳًﺎ ﻣﻦ ﺍﻟﻤﺘﻄﻠﺒﺎﺕ ﺍﻟﺮﻭﺗﻴﻨﻴﺔ‬.‫ﺍﻟﻨﻤﻮ ﺍﻟﺸﺎﻣﻠﺔ‬
(PDA) ‫ ﻟﻘﺪ ﺣﻈﻲ ﻣﺼﻄﻠﺢ ﺍﻟﻤﺴﺎﻋﺪ ﺍﻟﺮﻗﻤﻲ ﺍﻟﺸﺨﺼﻲ‬.‫ ﺃﻭ ﺍﻟﺸﻌﻮﺭ ﺑﺎﻟﺬﻧﺐ‬،‫ ﻭﺍﺣﺘﺮﺍﻡ ﺍﻟﺬﺍﺕ‬،‫ﻋﻠﻰ ﺗﺤﺪﻳﺪ ﺍﻟﻬﻮﻳﺔ ﺍﻻﺟﺘﻤﺎﻋﻴﺔ‬
[36].‫ﺑﺎﻫﺘﻤﺎﻡ ﻛﺒﻴﺮ ﻭﺃﺛﺎﺭ ﺟﺪﻻً ﻛﺒﻴﺮًﺍ ﻓﻲ ﺍﻷﻭﺳﺎﻁ ﺍﻷﻛﺎﺩﻳﻤﻴﺔ‬
PDA has garnered growing attention and has become a topic of discussion within the United
Kingdom. The term was first used by Elizabeth Newson during the 1980s to characterise children
who had widespread developmental problems and were brought to clinical attention due to their
pronounced inclination to evade routine demands and requests. The individuals in question used
ostensibly "manipulative" conduct as a means of evading requests, using actions that would be
considered startling or provocative by others. Additionally, the individuals exhibited abrupt
fluctuations in mood that were motivated by a need for dominance and control. They show a
deficiency in their perception of social identity, as well as a lack of pride or guilt. Furthermore,
they demonstrated a proclivity for engaging in role play and pretence, such as adopting the styles
and behaviours of others as a coping mechanism. Obsessions, including the avoidance of demands
and preoccupations with other individuals, were often seen. Additional characteristics observed in
the subjects were linguistic delays and the presence of "neurological soft signs," which included
delays in achieving motor milestones. Newson's descriptions elucidate the fact that children often
encounter significant difficulties within their domestic and educational environments, hence
engendering a sense of desperation among families who are in dire need of assistance. The pursuit
of regular demands has the potential to escalate a child's anxiety, resulting in manifestations of
verbal or physical aggressiveness, as well as the expression of threats to inflict harm against
oneself or others. In several households, the scheduling and planning of activities are centred on
meeting the needs and preferences of the kid. Efforts to actively mitigate circumstances in order to
minimise instances of meltdowns or hostile outbursts, so guaranteeing the child's continued
residence within the household, impose a significant weight on families. In the United Kingdom,
there has been a significant surge in interest PDA during the last decade, surpassing the
corresponding research efforts on this subject matter. The individuals who self-identify as adults
with PDA, parents of children diagnosed with PDA, and young individuals have played a crucial
6
‫‪role in the advancement of understanding about this condition.‬‬ ‫‪Nevertheless, the dearth of‬‬
‫‪research pertaining to PDA poses significant obstacles for therapists, as they are confronted with a‬‬
‫‪restricted foundation of empirical knowledge from which to derive their conclusions [37].‬‬

‫ﻟﻘﺪ ﺣﻈﻲ ﺍﻟﻤﺴﺎﻋﺪ ﺍﻟﺮﻗﻤﻲ ﺍﻟﺸﺨﺼﻲ ﺑﺎﻫﺘﻤﺎﻡ ﻣﺘﺰﺍﻳﺪ ﻭﺃﺻﺒﺢ ﻣﻮﺿﻮﻋًﺎ ﻟﻠﻨﻘﺎﺵ ﺩﺍﺧﻞ ﺍﻟﻤﻤﻠﻜﺔ ﺍﻟﻤﺘﺤﺪﺓ‪ .‬ﺗﻢ ﺍﺳﺘﺨﺪﺍﻡ ﻫﺬﺍ ﺍﻟﻤﺼﻄﻠﺢ‬
‫ﻷﻭﻝ ﻣﺮﺓ ﻣﻦ ﻗﺒﻞ ﺇﻟﻴﺰﺍﺑﻴﺚ ﻧﻴﻮﺳﻮﻥ ﺧﻼﻝ ﺍﻟﺜﻤﺎﻧﻴﻨﻴﺎﺕ ﻟﻮﺻﻒ ﺍﻷﻃﻔﺎﻝ ﺍﻟﺬﻳﻦ ﻳﻌﺎﻧﻮﻥ ﻣﻦ ﻣﺸﺎﻛﻞ ﻧﻤﻮ ﻭﺍﺳﻌﺔ ﺍﻟﻨﻄﺎﻕ ﻭﺗﻢ ﺗﻘﺪﻳﻤﻬﻢ‬
‫ﺇﻟﻰ ﺍﻟﺮﻋﺎﻳﺔ ﺍﻟﺴﺮﻳﺮﻳﺔ ﺑﺴﺒﺐ ﻣﻴﻠﻬﻢ ﺍﻟﻮﺍﺿﺢ ﻟﻠﺘﻬﺮﺏ ﻣﻦ ﺍﻟﻤﺘﻄﻠﺒﺎﺕ ﻭﺍﻟﻄﻠﺒﺎﺕ ﺍﻟﺮﻭﺗﻴﻨﻴﺔ‪ .‬ﺍﺳﺘﺨﺪﻡ ﺍﻷﻓﺮﺍﺩ ﺍﻟﻤﻌﻨﻴﻮﻥ ﺳﻠﻮﻛًﺎ‬
‫"ﻣﺘﻼﻋﺒًﺎ" ﻇﺎﻫﺮﻳًﺎ ﻛﻮﺳﻴﻠﺔ ﻟﻠﺘﻬﺮﺏ ﻣﻦ ﺍﻟﻄﻠﺒﺎﺕ‪ ،‬ﻭﺫﻟﻚ ﺑﺎﺳﺘﺨﺪﺍﻡ ﺃﻓﻌﺎﻝ ﻗﺪ ﻳﻌﺘﺒﺮﻫﺎ ﺍﻵﺧﺮﻭﻥ ﻣﺬﻫﻠﺔ ﺃﻭ ﺍﺳﺘﻔﺰﺍﺯﻳﺔ‪ .‬ﺑﺎﻹﺿﺎﻓﺔ ﺇﻟﻰ‬
‫ﺫﻟﻚ‪ ،‬ﺃﻇﻬﺮ ﺍﻷﻓﺮﺍﺩ ﺗﻘﻠﺒﺎﺕ ﻣﻔﺎﺟﺌﺔ ﻓﻲ ﺍﻟﺤﺎﻟﺔ ﺍﻟﻤﺰﺍﺟﻴﺔ ﻛﺎﻧﺖ ﻣﺪﻓﻮﻋﺔ ﺑﺎﻟﺤﺎﺟﺔ ﺇﻟﻰ ﺍﻟﻬﻴﻤﻨﺔ ﻭﺍﻟﺴﻴﻄﺮﺓ‪ .‬ﻭﻳﻈﻬﺮﻭﻥ ﻧﻘﺼًﺎ ﻓﻲ‬
‫ﺇﺩﺭﺍﻛﻬﻢ ﻟﻠﻬﻮﻳﺔ ﺍﻻﺟﺘﻤﺎﻋﻴﺔ‪ ،‬ﻓﻀﻼً ﻋﻦ ﻋﺪﻡ ﻭﺟﻮﺩ ﻓﺨﺮ ﺃﻭ ﺷﻌﻮﺭ ﺑﺎﻟﺬﻧﺐ‪ .‬ﻋﻼﻭﺓ ﻋﻠﻰ ﺫﻟﻚ‪ ،‬ﻓﻘﺪ ﺃﻇﻬﺮﻭﺍ ﻣﻴﻠًﺎ ﻟﻼﻧﺨﺮﺍﻁ ﻓﻲ ﻟﻌﺐ‬
‫ﺍﻷﺩﻭﺍﺭ ﻭﺍﻟﺘﻈﺎﻫﺮ‪ ،‬ﻣﺜﻞ ﺗﺒﻨﻲ ﺃﺳﺎﻟﻴﺐ ﻭﺳﻠﻮﻛﻴﺎﺕ ﺍﻵﺧﺮﻳﻦ ﻛﺂﻟﻴﺔ ﻟﻠﺘﻜﻴﻒ‪ .‬ﻭﻛﺜﻴﺮﺍ ﻣﺎ ﺷﻮﻫﺪﺕ ﺍﻟﻬﻮﺍﺟﺲ‪ ،‬ﺑﻤﺎ ﻓﻲ ﺫﻟﻚ ﺗﺠﻨﺐ‬
‫ﺍﻟﻤﻄﺎﻟﺐ ﻭﺍﻻﻧﺸﻐﺎﻝ ﺑﺎﻷﻓﺮﺍﺩ ﺍﻵﺧﺮﻳﻦ‪ .‬ﻭﻛﺎﻧﺖ ﺍﻟﺨﺼﺎﺋﺺ ﺍﻹﺿﺎﻓﻴﺔ ﺍﻟﺘﻲ ﻟﻮﺣﻈﺖ ﻓﻲ ﺍﻟﻤﻮﺍﺿﻴﻊ ﻫﻲ ﺍﻟﺘﺄﺧﺮ ﺍﻟﻠﻐﻮﻱ ﻭﻭﺟﻮﺩ‬
‫"ﻋﻼﻣﺎﺕ ﻋﺼﺒﻴﺔ ﻧﺎﻋﻤﺔ"‪ ،‬ﻭﺍﻟﺘﻲ ﺗﻀﻤﻨﺖ ﺍﻟﺘﺄﺧﺮ ﻓﻲ ﺗﺤﻘﻴﻖ ﺍﻟﻤﻌﺎﻟﻢ ﺍﻟﺤﺮﻛﻴﺔ‪ .‬ﺗﻮﺿﺢ ﺃﻭﺻﺎﻑ ﻧﻴﻮﺳﻮﻥ ﺣﻘﻴﻘﺔ ﺃﻥ ﺍﻷﻃﻔﺎﻝ ﻏﺎﻟﺒًﺎ ﻣﺎ‬
‫ﻳﻮﺍﺟﻬﻮﻥ ﺻﻌﻮﺑﺎﺕ ﻛﺒﻴﺮﺓ ﺩﺍﺧﻞ ﺑﻴﺌﺎﺗﻬﻢ ﺍﻟﻤﻨﺰﻟﻴﺔ ﻭﺍﻟﺘﻌﻠﻴﻤﻴﺔ‪ ،‬ﻣﻤﺎ ﻳﻮﻟﺪ ﺷﻌﻮﺭًﺍ ﺑﺎﻟﻴﺄﺱ ﺑﻴﻦ ﺍﻟﻌﺎﺋﻼﺕ ﺍﻟﺘﻲ ﻫﻲ ﻓﻲ ﺣﺎﺟﺔ ﻣﺎﺳﺔ ﺇﻟﻰ‬
‫ﺍﻟﻤﺴﺎﻋﺪﺓ‪ .‬ﺇﻥ ﺍﻟﺴﻌﻲ ﻟﺘﺤﻘﻴﻖ ﻣﻄﺎﻟﺐ ﻣﻨﺘﻈﻤﺔ ﻣﻦ ﺷﺄﻧﻪ ﺃﻥ ﻳﺆﺩﻱ ﺇﻟﻰ ﺗﺼﻌﻴﺪ ﺍﻟﻘﻠﻖ ﻟﺪﻯ ﺍﻟﻄﻔﻞ‪ ،‬ﻣﻤﺎ ﻳﺆﺩﻱ ﺇﻟﻰ ﻇﻬﻮﺭ ﻣﻈﺎﻫﺮ‬
‫ﺍﻟﻌﺪﻭﺍﻥ ﺍﻟﻠﻔﻈﻲ ﺃﻭ ﺍﻟﺠﺴﺪﻱ‪ ،‬ﻓﻀﻼً ﻋﻦ ﺍﻟﺘﻌﺒﻴﺮ ﻋﻦ ﺍﻟﺘﻬﺪﻳﺪﺍﺕ ﺑﺈﻟﺤﺎﻕ ﺍﻷﺫﻯ ﺑﺎﻟﻨﻔﺲ ﺃﻭ ﺑﺎﻵﺧﺮﻳﻦ‪ .‬ﻓﻲ ﺍﻟﻌﺪﻳﺪ ﻣﻦ ﺍﻷﺳﺮ‪ ،‬ﺗﺮﻛﺰ‬
‫ﺟﺪﻭﻟﺔ ﻭﺗﺨﻄﻴﻂ ﺍﻷﻧﺸﻄﺔ ﻋﻠﻰ ﺗﻠﺒﻴﺔ ﺍﺣﺘﻴﺎﺟﺎﺕ ﺍﻟﻄﻔﻞ ﻭﺗﻔﻀﻴﻼﺗﻪ‪ .‬ﺇﻥ ﺍﻟﺠﻬﻮﺩ ﺍﻟﻤﺒﺬﻭﻟﺔ ﻟﻠﺘﺨﻔﻴﻒ ﻣﻦ ﺍﻟﻈﺮﻭﻑ ﺑﺸﻜﻞ ﻓﻌﺎﻝ ﻣﻦ ﺃﺟﻞ‬
‫ﺗﻘﻠﻴﻞ ﺣﺎﻻﺕ ﺍﻻﻧﻬﻴﺎﺭﺍﺕ ﺃﻭ ﺍﻻﻧﻔﺠﺎﺭﺍﺕ ﺍﻟﻌﺪﺍﺋﻴﺔ‪ ،‬ﻭﺑﺎﻟﺘﺎﻟﻲ ﻓﺈﻥ ﺿﻤﺎﻥ ﺍﺳﺘﻤﺮﺍﺭ ﺇﻗﺎﻣﺔ ﺍﻟﻄﻔﻞ ﺩﺍﺧﻞ ﺍﻷﺳﺮﺓ‪ ،‬ﻳﻔﺮﺽ ﺛﻘﻠًﺎ ﻛﺒﻴﺮًﺍ ﻋﻠﻰ‬
‫ﺍﻟﻌﺎﺋﻼﺕ‪ .‬ﻓﻲ ﺍﻟﻤﻤﻠﻜﺔ ﺍﻟﻤﺘﺤﺪﺓ‪ ،‬ﻛﺎﻥ ﻫﻨﺎﻙ ﺍﺭﺗﻔﺎﻉ ﻛﺒﻴﺮ ﻓﻲ ﺍﻻﻫﺘﻤﺎﻡ ﺑﺎﻟﻤﺴﺎﻋﺪ ﺍﻟﺸﺨﺼﻲ ﺍﻟﺮﻗﻤﻲ ﺧﻼﻝ ﺍﻟﻌﻘﺪ ﺍﻟﻤﺎﺿﻲ‪ ،‬ﻣﺘﺠﺎﻭﺯًﺍ‬
‫ﺍﻟﺠﻬﻮﺩ ﺍﻟﺒﺤﺜﻴﺔ ﺍﻟﻤﻘﺎﺑﻠﺔ ﺣﻮﻝ ﻫﺬﺍ ﺍﻟﻤﻮﺿﻮﻉ‪ .‬ﻟﻘﺪ ﻟﻌﺐ ﺍﻷﻓﺮﺍﺩ ﺍﻟﺬﻳﻦ ﻳﻌﺮﻓﻮﻥ ﺃﻧﻔﺴﻬﻢ ﻛﺒﺎﻟﻐﻴﻦ ﻣﺼﺎﺑﻴﻦ ﺑﺎﻟﻘﻨﺎﺓ ﺍﻟﺸﺮﻳﺎﻧﻴﺔ ﺍﻟﺴﺎﻟﻜﺔ‪،‬‬
‫ﻭﺃﻭﻟﻴﺎﺀ ﺃﻣﻮﺭ ﺍﻷﻃﻔﺎﻝ ﺍﻟﺬﻳﻦ ﺗﻢ ﺗﺸﺨﻴﺺ ﺇﺻﺎﺑﺘﻬﻢ ﺑﺎﻟﻘﻨﺎﺓ ﺍﻟﺸﺮﻳﺎﻧﻴﺔ ﺍﻟﺴﺎﻟﻜﺔ‪ ،‬ﻭﺍﻟﺸﺒﺎﺏ ﺩﻭﺭًﺍ ﺣﺎﺳﻤًﺎ ﻓﻲ ﺗﻌﺰﻳﺰ ﻓﻬﻢ ﻫﺬﻩ ﺍﻟﺤﺎﻟﺔ‪ .‬ﻭﻣﻊ‬
‫ﺫﻟﻚ‪ ،‬ﻓﺈﻥ ﻧﺪﺭﺓ ﺍﻷﺑﺤﺎﺙ ﺍﻟﻤﺘﻌﻠﻘﺔ ﺑﺎﻟﻤﺴﺎﻋﺪ ﺍﻟﺸﺨﺼﻲ ﺍﻟﺮﻗﻤﻲ ﺗﺸﻜﻞ ﻋﻘﺒﺎﺕ ﻛﺒﻴﺮﺓ ﺃﻣﺎﻡ ﺍﻟﻤﻌﺎﻟﺠﻴﻦ‪ ،‬ﺣﻴﺚ ﻳﻮﺍﺟﻬﻮﻥ ﺃﺳﺎﺳًﺎ ﻣﻘﻴﺪًﺍ‬
‫ﻟﻠﻤﻌﺮﻓﺔ ﺍﻟﺘﺠﺮﻳﺒﻴﺔ ﺍﻟﺘﻲ ﻳﻤﻜﻨﻬﻢ ﻣﻦ ﺧﻼﻟﻬﺎ ﺍﺳﺘﺨﻼﺹ ﺍﺳﺘﻨﺘﺎﺟﺎﺗﻬﻢ ]‪.[37‬‬

‫‪2.3.3 Childhood Disintegrative Disorder‬‬

‫‪The first documentation of Childhood Disintegrative Disorder (CDD) was provided by‬‬
‫‪Theodore Heller. It is a rare condition has a low occurrence rate, affecting around 1 in every‬‬
‫‪100,000 children. The disorder was first referred to as dementia infantilis, and has since been‬‬
‫‪alternatively referred to as disintegrative psychosis or Heller's syndrome. This illness has a‬‬
‫‪7‬‬
multifaceted nature, impacting several domains of a child's developmental trajectory. This
syndrome is distinguished by an initial period of typical development lasting at least two years
after birth. During this time, individuals exhibit adequate levels of verbal and nonverbal
communication, engage in social interactions, participate in play activities, and demonstrate
adaptive behaviour. The onset of the condition often occurs between the ages of 3 and 4, however
it may manifest at any point up to the age of 10. [38].

‫ ﺍﻻﺿﻄﺮﺍﺏ ﺍﻻﻧﺤﻼﻟﻲ ﻓﻲ ﻣﺮﺣﻠﺔ ﺍﻟﻄﻔﻮﻟﺔ‬2.3.3

،‫ ﻭﻫﻲ ﺣﺎﻟﺔ ﻧﺎﺩﺭﺓ ﺫﺍﺕ ﻣﻌﺪﻝ ﺣﺪﻭﺙ ﻣﻨﺨﻔﺾ‬.‫( ﺑﻮﺍﺳﻄﺔ ﺛﻴﻮﺩﻭﺭ ﻫﻴﻠﺮ‬CDD) ‫ﺗﻢ ﺗﻘﺪﻳﻢ ﺃﻭﻝ ﺗﻮﺛﻴﻖ ﻻﺿﻄﺮﺍﺏ ﺍﻟﻄﻔﻮﻟﺔ ﺍﻟﺘﻔﻜﻜﻲ‬
‫ ﻭﻣﻨﺬ‬،‫ ﺗﻤﺖ ﺍﻹﺷﺎﺭﺓ ﺇﻟﻰ ﻫﺬﺍ ﺍﻻﺿﻄﺮﺍﺏ ﻓﻲ ﺍﻟﺒﺪﺍﻳﺔ ﺑﺎﺳﻢ ﺍﻟﺨﺮﻑ ﺍﻟﻄﻔﻠﻲ‬.‫ ﻃﻔﻞ‬100.000 ‫ ﻣﻦ ﻛﻞ‬1 ‫ﺣﻴﺚ ﺗﺆﺛﺮ ﻋﻠﻰ ﺣﻮﺍﻟﻲ‬
‫ ﻭﻳﺆﺛﺮ ﻋﻠﻰ ﻋﺪﺓ‬،‫ ﻫﺬﺍ ﺍﻟﻤﺮﺽ ﻟﻪ ﻃﺒﻴﻌﺔ ﻣﺘﻌﺪﺩﺓ ﺍﻷﻭﺟﻪ‬.‫ﺫﻟﻚ ﺍﻟﺤﻴﻦ ﺗﻤﺖ ﺍﻹﺷﺎﺭﺓ ﺇﻟﻴﻪ ﺑﺎﺳﻢ ﺍﻟﺬﻫﺎﻥ ﺍﻟﺘﺤﻠﻠﻲ ﺃﻭ ﻣﺘﻼﺯﻣﺔ ﻫﻴﻠﺮ‬
‫ ﺗﺘﻤﻴﺰ ﻫﺬﻩ ﺍﻟﻤﺘﻼﺯﻣﺔ ﺑﻔﺘﺮﺓ ﺃﻭﻟﻴﺔ ﻣﻦ ﺍﻟﺘﻄﻮﺭ ﺍﻟﻨﻤﻮﺫﺟﻲ ﺗﺴﺘﻤﺮ ﻟﻤﺪﺓ ﻋﺎﻣﻴﻦ ﻋﻠﻰ ﺍﻷﻗﻞ ﺑﻌﺪ‬.‫ﻣﺠﺎﻻﺕ ﻣﻦ ﻣﺴﺎﺭ ﻧﻤﻮ ﺍﻟﻄﻔﻞ‬
‫ ﻭﻳﻨﺨﺮﻃﻮﻥ ﻓﻲ ﺍﻟﺘﻔﺎﻋﻼﺕ‬،‫ ﻳﻈﻬﺮ ﺍﻷﻓﺮﺍﺩ ﻣﺴﺘﻮﻳﺎﺕ ﻛﺎﻓﻴﺔ ﻣﻦ ﺍﻟﺘﻮﺍﺻﻞ ﺍﻟﻠﻔﻈﻲ ﻭﻏﻴﺮ ﺍﻟﻠﻔﻈﻲ‬،‫ ﺧﻼﻝ ﻫﺬﺍ ﺍﻟﻮﻗﺖ‬.‫ﺍﻟﻮﻻﺩﺓ‬
‫ ﻭﻣﻊ‬،‫ ﺳﻨﻮﺍﺕ‬4‫ ﻭ‬3 ‫ ﻏﺎﻟﺒًﺎ ﻣﺎ ﺗﺤﺪﺙ ﺑﺪﺍﻳﺔ ﺍﻟﺤﺎﻟﺔ ﺑﻴﻦ ﺳﻦ‬.‫ ﻭﻳﻈﻬﺮﻭﻥ ﺍﻟﺴﻠﻮﻙ ﺍﻟﺘﻜﻴﻔﻲ‬،‫ ﻭﻳﺸﺎﺭﻛﻮﻥ ﻓﻲ ﺃﻧﺸﻄﺔ ﺍﻟﻠﻌﺐ‬،‫ﺍﻻﺟﺘﻤﺎﻋﻴﺔ‬
[38] .‫ ﺳﻨﻮﺍﺕ‬10 ‫ﺫﻟﻚ ﻗﺪ ﺗﻈﻬﺮ ﻓﻲ ﺃﻱ ﻭﻗﺖ ﺣﺘﻰ ﺳﻦ‬
CDD, is a very uncommon and poorly understood variant of Pervasive developmental disorders
PDD. CDD is characterised by a phase of typical developmental progression in children lasting a
minimum of two years, followed by a subsequent phase of significant and permanent decline in
both social and cognitive abilities. Following the regression analysis, it can be shown that the
behavioural characteristics exhibited by individuals diagnosed with CDD have resemblance to
those often associated with ASD. Nevertheless, the seeming regular progression of development
in individuals with CDD prior to the commencement of regression, as well as the occurrence of
regression at a later age, are often regarded as characteristic features of this illness and indicative
of its differentiation from other diagnoses within the ASD. The scarcity of CDD has posed
significant difficulties in doing research on this disorder, resulting in a lack of clarity on its
diagnostic features and even its diagnostic reliability[39].

‫ ﻳﺘﻤﻴﺰ ﺍﺿﻄﺮﺍﺏ ﺍﻟﺸﺨﺼﻴﺔ ﺍﻟﺘﻨﻤﻮﻳﺔ‬.(PDD) ‫ ﻫﻮ ﺷﻜﻞ ﻏﻴﺮ ﺷﺎﺋﻊ ﻭﻏﻴﺮ ﻣﻔﻬﻮﻡ ﺟﻴﺪًﺍ ﻣﻦ ﺍﺿﻄﺮﺍﺑﺎﺕ ﺍﻟﻨﻤﻮ ﺍﻟﺸﺎﻣﻠﺔ‬،CDD
‫ ﺗﻠﻴﻬﺎ ﻣﺮﺣﻠﺔ ﻻﺣﻘﺔ ﻣﻦ ﺍﻻﻧﺨﻔﺎﺽ ﺍﻟﻜﺒﻴﺮ‬،‫ﺑﻤﺮﺣﻠﺔ ﻣﻦ ﺍﻟﺘﻘﺪﻡ ﺍﻟﺘﻨﻤﻮﻱ ﺍﻟﻨﻤﻮﺫﺟﻲ ﻟﺪﻯ ﺍﻷﻃﻔﺎﻝ ﺗﺴﺘﻤﺮ ﻟﻤﺪﺓ ﻋﺎﻣﻴﻦ ﻋﻠﻰ ﺍﻷﻗﻞ‬
‫ ﻳﻤﻜﻦ ﺇﺛﺒﺎﺕ ﺃﻥ ﺍﻟﺨﺼﺎﺋﺺ ﺍﻟﺴﻠﻮﻛﻴﺔ ﺍﻟﺘﻲ ﻳﻈﻬﺮﻫﺎ‬،‫ ﺑﻌﺪ ﺗﺤﻠﻴﻞ ﺍﻻﻧﺤﺪﺍﺭ‬.‫ﻭﺍﻟﺪﺍﺋﻢ ﻓﻲ ﻛﻞ ﻣﻦ ﺍﻟﻘﺪﺭﺍﺕ ﺍﻻﺟﺘﻤﺎﻋﻴﺔ ﻭﺍﻟﻤﻌﺮﻓﻴﺔ‬

8
‫ ﻓﺈﻥ ﺍﻟﺘﻘﺪﻡ ﺍﻟﻤﻨﺘﻈﻢ‬،‫ ﻭﻣﻊ ﺫﻟﻚ‬.‫ ﺗﺸﺒﻪ ﺗﻠﻚ ﺍﻟﻤﺮﺗﺒﻄﺔ ﻏﺎﻟﺒًﺎ ﺑﺎﺿﻄﺮﺍﺏ ﻃﻴﻒ ﺍﻟﺘﻮﺣﺪ‬CDD ‫ﺍﻷﻓﺮﺍﺩ ﺍﻟﺬﻳﻦ ﺗﻢ ﺗﺸﺨﻴﺼﻬﻢ ﺑﻤﺮﺽ‬
،‫ ﻭﻛﺬﻟﻚ ﺣﺪﻭﺙ ﺍﻟﺘﺮﺍﺟﻊ ﻓﻲ ﺳﻦ ﻻﺣﻘﺔ‬،‫ﺍﻟﻈﺎﻫﺮ ﻓﻲ ﺍﻟﻨﻤﻮ ﻟﺪﻯ ﺍﻷﻓﺮﺍﺩ ﺍﻟﻤﺼﺎﺑﻴﻦ ﺑﺎﺿﻄﺮﺍﺑﺎﺕ ﺍﻟﺸﺨﺼﻴﺔ ﺍﻟﻤﺰﻣﻨﺔ ﻗﺒﻞ ﺑﺪﺀ ﺍﻟﺘﺮﺍﺟﻊ‬
.‫ﻏﺎﻟﺒًﺎ ﻣﺎ ﻳُﻨﻈﺮ ﺇﻟﻴﻬﻢ ﻋﻠﻰ ﺃﻧﻬﻢ ﺳﻤﺎﺕ ﻣﻤﻴﺰﺓ ﻟﻬﺬﺍ ﺍﻟﻤﺮﺽ ﻭﻣﺆﺷﺮ ﻋﻠﻰ ﺗﻤﺎﻳﺰﻩ ﻋﻦ ﺍﻟﺘﺸﺨﻴﺼﺎﺕ ﺍﻷﺧﺮﻯ ﺩﺍﺧﻞ ﺍﻟﻤﺠﺘﻤﻊ‬
‫ ﻣﻤﺎ ﺃﺩﻯ ﺇﻟﻰ ﻋﺪﻡ‬،‫ ﺻﻌﻮﺑﺎﺕ ﻛﺒﻴﺮﺓ ﻓﻲ ﺇﺟﺮﺍﺀ ﺍﻷﺑﺤﺎﺙ ﺣﻮﻝ ﻫﺬﺍ ﺍﻻﺿﻄﺮﺍﺏ‬CDD ‫ ﻟﻘﺪ ﺷﻜﻠﺖ ﻧﺪﺭﺓ‬.‫ﺍﺿﻄﺮﺍﺏ ﻃﻴﻒ ﺍﻟﺘﻮﺣﺪ‬
[39] ‫ﺍﻟﻮﺿﻮﺡ ﺑﺸﺄﻥ ﻣﻴﺰﺍﺗﻪ ﺍﻟﺘﺸﺨﻴﺼﻴﺔ ﻭﺣﺘﻰ ﻣﻮﺛﻮﻗﻴﺔ ﺍﻟﺘﺸﺨﻴﺺ‬
2.3.4 Pervasive Developmental Disorder
Pervasive developmental disorders (PDD) are distinguished by impairments in the acquisition of
social and communicative abilities. Symptoms of the condition may be seen by parents throughout
infancy, however the average age at which they manifest is about 3 years.[40]

‫ ﺍﺿﻄﺮﺍﺏ ﺍﻟﻨﻤﻮ ﺍﻟﻤﻨﺘﺸﺮ‬2.3.4 .

‫ ﻗﺪ ﺗﻈﻬﺮ ﺃﻋﺮﺍﺽ ﻫﺬﻩ ﺍﻟﺤﺎﻟﺔ ﻋﻠﻰ‬.‫( ﺑﻀﻌﻒ ﻓﻲ ﺍﻛﺘﺴﺎﺏ ﺍﻟﻘﺪﺭﺍﺕ ﺍﻻﺟﺘﻤﺎﻋﻴﺔ ﻭﺍﻟﺘﻮﺍﺻﻠﻴﺔ‬PDD) ‫ﺗﺘﻤﻴﺰ ﺍﺿﻄﺮﺍﺑﺎﺕ ﺍﻟﻨﻤﻮ ﺍﻟﺸﺎﻣﻠﺔ‬
[40].‫ ﺳﻨﻮﺍﺕ‬3 ‫ ﻭﻟﻜﻦ ﻣﺘﻮﺳﻂ ﺍﻟﻌﻤﺮ ﺍﻟﺬﻱ ﺗﻈﻬﺮ ﻓﻴﻪ ﺍﻷﻋﺮﺍﺽ ﻫﻮ ﺣﻮﺍﻟﻲ‬،‫ﺍﻟﻮﺍﻟﺪﻳﻦ ﻃﻮﺍﻝ ﻣﺮﺣﻠﺔ ﺍﻟﻄﻔﻮﻟﺔ‬
Among the symptoms might be difficulties in speaking and comprehending language, inability to
relate to people, things, or occasions, Various ways to play with toys and other items, Inability to
adapt to routine or environment changes and repetitive gestures or habits of behavior [41].

‫ ﻭﻋﺪﻡ ﺍﻟﻘﺪﺭﺓ ﻋﻠﻰ ﺍﻟﺘﻮﺍﺻﻞ ﻣﻊ ﺍﻷﺷﺨﺎﺹ ﺃﻭ ﺍﻷﺷﻴﺎﺀ ﺃﻭ‬،‫ﻗﺪ ﻳﻜﻮﻥ ﻣﻦ ﺑﻴﻦ ﺍﻷﻋﺮﺍﺽ ﺻﻌﻮﺑﺎﺕ ﻓﻲ ﺍﻟﺘﺤﺪﺙ ﻭﻓﻬﻢ ﺍﻟﻠﻐﺔ‬
‫ ﻋﺪﻡ ﺍﻟﻘﺪﺭﺓ ﻋﻠﻰ ﺍﻟﺘﻜﻴﻒ ﻣﻊ ﺍﻟﺘﻐﻴﺮﺍﺕ ﺍﻟﺮﻭﺗﻴﻨﻴﺔ ﺃﻭ ﺍﻟﺒﻴﺌﻴﺔ ﻭﺍﻹﻳﻤﺎﺀﺍﺕ‬،‫ ﻃﺮﻕ ﻣﺨﺘﻠﻔﺔ ﻟﻠﻌﺐ ﺑﺎﻷﻟﻌﺎﺏ ﻭﺍﻷﺷﻴﺎﺀ ﺍﻷﺧﺮﻯ‬،‫ﺍﻟﻤﻨﺎﺳﺒﺎﺕ‬
.[ [41] ‫ﺍﻟﻤﺘﻜﺮﺭﺓ ﺃﻭ ﻋﺎﺩﺍﺕ ﺍﻟﺴﻠﻮﻙ‬
There is considerable variability among children diagnosed with PDD. There exists a subset of
children that exhibit a complete absence of speech, whilst others demonstrate restricted use of
phrases or dialogues, and a third group displays language development that falls within the usual
range. Typically, individuals with repetitive play skills and low social abilities have observable
characteristics. Exaggerated reactions to sensory stimuli, such as high decibel levels and intense
illumination, are also prevalent. Currently, there is a lack of a definitive remedy for PDD.
Pharmaceutical interventions might be used to target distinct behavioural issues, while therapeutic
approaches have to be tailored to the unique requirements of individual children. Certain children
diagnosed with PDD may have positive outcomes when placed in specialised classrooms, while
others may thrive in ordinary special education courses or mainstream classrooms with
9
‫‪supplementary assistance. The use of early intervention strategies is of utmost importance in‬‬
‫]‪enhancing the overall prognosis of persons diagnosed with PDD[40‬‬

‫ﻫﻨﺎﻙ ﺗﺒﺎﻳﻦ ﻛﺒﻴﺮ ﺑﻴﻦ ﺍﻷﻃﻔﺎﻝ ﺍﻟﺬﻳﻦ ﺗﻢ ﺗﺸﺨﻴﺺ ﺇﺻﺎﺑﺘﻬﻢ ﺑﺎﻻﺿﻄﺮﺍﺑﺎﺕ ﺍﻟﻨﻤﺎﺋﻴﺔ ﺍﻟﺸﺎﻣﻠﺔ‪ .‬ﻫﻨﺎﻙ ﻣﺠﻤﻮﻋﺔ ﻓﺮﻋﻴﺔ ﻣﻦ ﺍﻷﻃﻔﺎﻝ‬
‫ﻳﻈﻬﺮﻭﻥ ﻏﻴﺎﺑًﺎ ﺗﺎﻣًﺎ ﻟﻠﻜﻼﻡ‪ ،‬ﺑﻴﻨﻤﺎ ﻳﻈﻬﺮ ﺁﺧﺮﻭﻥ ﺍﺳﺘﺨﺪﺍﻣًﺎ ﻣﻘﻴﺪًﺍ ﻟﻠﻌﺒﺎﺭﺍﺕ ﺃﻭ ﺍﻟﺤﻮﺍﺭﺍﺕ‪ ،‬ﻭﺗﻈﻬﺮ ﻣﺠﻤﻮﻋﺔ ﺛﺎﻟﺜﺔ ﺗﻄﻮﺭ ﺍﻟﻠﻐﺔ ﺍﻟﺬﻱ‬
‫ﻳﻘﻊ ﺿﻤﻦ ﺍﻟﻨﻄﺎﻕ ﺍﻟﻤﻌﺘﺎﺩ‪ .‬ﻋﺎﺩﺓً ﻣﺎ ﻳﻜﻮﻥ ﻟﺪﻯ ﺍﻷﻓﺮﺍﺩ ﺫﻭﻱ ﻣﻬﺎﺭﺍﺕ ﺍﻟﻠﻌﺐ ﺍﻟﻤﺘﻜﺮﺭﺓ ﻭﺍﻟﻘﺪﺭﺍﺕ ﺍﻻﺟﺘﻤﺎﻋﻴﺔ ﺍﻟﻤﻨﺨﻔﻀﺔ ﺧﺼﺎﺋﺺ‬
‫ﻳﻤﻜﻦ ﻣﻼﺣﻈﺘﻬﺎ‪ .‬ﻛﻤﺎ ﺃﻥ ﺭﺩﻭﺩ ﺍﻟﻔﻌﻞ ﺍﻟﻤﺒﺎﻟﻎ ﻓﻴﻬﺎ ﺗﺠﺎﻩ ﺍﻟﻤﺤﻔﺰﺍﺕ ﺍﻟﺤﺴﻴﺔ‪ ،‬ﻣﺜﻞ ﻣﺴﺘﻮﻳﺎﺕ ﺍﻟﺪﻳﺴﻴﺒﻞ ﺍﻟﻌﺎﻟﻴﺔ ﻭﺍﻹﺿﺎﺀﺓ ﺍﻟﺸﺪﻳﺪﺓ‪،‬‬
‫ﺳﺎﺋﺪﺓ ﺃﻳﻀًﺎ‪ .‬ﻓﻲ ﺍﻟﻮﻗﺖ ﺍﻟﺤﺎﻟﻲ‪ ،‬ﻻ ﻳﻮﺟﺪ ﻋﻼﺝ ﻧﻬﺎﺋﻲ ﻟﻼﺿﻄﺮﺍﺑﺎﺕ ﺍﻟﻨﻤﺎﺋﻴﺔ ﺍﻟﺸﺎﻣﻠﺔ‪ .‬ﻳﻤﻜﻦ ﺍﺳﺘﺨﺪﺍﻡ ﺍﻟﺘﺪﺧﻼﺕ ﺍﻟﺼﻴﺪﻻﻧﻴﺔ‬
‫ﻻﺳﺘﻬﺪﺍﻑ ﻗﻀﺎﻳﺎ ﺳﻠﻮﻛﻴﺔ ﻣﺘﻤﻴﺰﺓ‪ ،‬ﻓﻲ ﺣﻴﻦ ﻳﺠﺐ ﺗﺼﻤﻴﻢ ﺍﻷﺳﺎﻟﻴﺐ ﺍﻟﻌﻼﺟﻴﺔ ﻭﻓﻘًﺎ ﻟﻠﻤﺘﻄﻠﺒﺎﺕ ﺍﻟﻔﺮﻳﺪﺓ ﻟﻜﻞ ﻃﻔﻞ ﻋﻠﻰ ﺣﺪﺓ‪ .‬ﻗﺪ ﻳﻜﻮﻥ‬
‫ﻟﺒﻌﺾ ﺍﻷﻃﻔﺎﻝ ﺍﻟﺬﻳﻦ ﺗﻢ ﺗﺸﺨﻴﺺ ﺇﺻﺎﺑﺘﻬﻢ ﺑﺎﻻﺿﻄﺮﺍﺑﺎﺕ ﺍﻟﻨﻤﺎﺋﻴﺔ ﺍﻟﺸﺎﻣﻠﺔ ﻧﺘﺎﺋﺞ ﺇﻳﺠﺎﺑﻴﺔ ﻋﻨﺪ ﻭﺿﻌﻬﻢ ﻓﻲ ﻓﺼﻮﻝ ﺩﺭﺍﺳﻴﺔ‬
‫ﻣﺘﺨﺼﺼﺔ‪ ،‬ﺑﻴﻨﻤﺎ ﻗﺪ ﻳﺰﺩﻫﺮ ﺁﺧﺮﻭﻥ ﻓﻲ ﺩﻭﺭﺍﺕ ﺍﻟﺘﻌﻠﻴﻢ ﺍﻟﺨﺎﺹ ﺍﻟﻌﺎﺩﻳﺔ ﺃﻭ ﺍﻟﻔﺼﻮﻝ ﺍﻟﺪﺭﺍﺳﻴﺔ ﺍﻟﻌﺎﺩﻳﺔ ﺑﻤﺴﺎﻋﺪﺓ ﺗﻜﻤﻴﻠﻴﺔ‪ .‬ﻳﻌﺪ‬
‫ﺍﺳﺘﺨﺪﺍﻡ ﺍﺳﺘﺮﺍﺗﻴﺠﻴﺎﺕ ﺍﻟﺘﺪﺧﻞ ﺍﻟﻤﺒﻜﺮ ﺫﺍ ﺃﻫﻤﻴﺔ ﻗﺼﻮﻯ ﻓﻲ ﺗﻌﺰﻳﺰ ﺍﻟﺘﺸﺨﻴﺺ ﺍﻟﻌﺎﻡ ﻟﻸﺷﺨﺎﺹ ﺍﻟﺬﻳﻦ ﺗﻢ ﺗﺸﺨﻴﺺ ﺇﺻﺎﺑﺘﻬﻢ‬
‫ﺑﺎﻻﺿﻄﺮﺍﺑﺎﺕ ﺍﻟﻨﻤﺎﺋﻴﺔ ﺍﻟﺸﺎﻣﻠﺔ ]‪.[40‬‬

‫‪2.4 Preprocessing‬‬
‫‪Preprocessing is a crucial preliminary stage in data processing that precedes the use of machine‬‬
‫‪learning techniques. The preprocessing of data is an essential step in the data analysis process,‬‬
‫‪since it involves the removal and extraction of irrelevant or undesired data. This is done to‬‬
‫‪enhance the efficiency and effectiveness of machine learning algorithms. Efficient performance of‬‬
‫‪machine learning (ML) algorithms is contingent upon the appropriate preprocessing of the‬‬
‫‪dataset[42]. This thesis employs several preprocessing approaches, including data balance, data‬‬
‫‪cleaning and label encoding. The subsequent subsections provide a comprehensive explanation of‬‬
‫‪these preprocessing techniques.‬‬

‫‪ 2.4‬ﺍﻟﻤﻌﺎﻟﺠﺔ ﺍﻟﻤﺴﺒﻘﺔ‬

‫‪10‬‬
‫ﺗﻌﺪ ﺍﻟﻤﻌﺎﻟﺠﺔ ﺍﻟﻤﺴﺒﻘﺔ ﻣﺮﺣﻠﺔ ﺃﻭﻟﻴﺔ ﺣﺎﺳﻤﺔ ﻓﻲ ﻣﻌﺎﻟﺠﺔ ﺍﻟﺒﻴﺎﻧﺎﺕ ﺍﻟﺘﻲ ﺗﺴﺒﻖ ﺍﺳﺘﺨﺪﺍﻡ ﺗﻘﻨﻴﺎﺕ ﺍﻟﺘﻌﻠﻢ ﺍﻵﻟﻲ‪ .‬ﺗﻌﺪ ﺍﻟﻤﻌﺎﻟﺠﺔ ﺍﻟﻤﺴﺒﻘﺔ‬
‫ﻟﻠﺒﻴﺎﻧﺎﺕ ﺧﻄﻮﺓ ﺃﺳﺎﺳﻴﺔ ﻓﻲ ﻋﻤﻠﻴﺔ ﺗﺤﻠﻴﻞ ﺍﻟﺒﻴﺎﻧﺎﺕ‪ ،‬ﻷﻧﻬﺎ ﺗﻨﻄﻮﻱ ﻋﻠﻰ ﺇﺯﺍﻟﺔ ﻭﺍﺳﺘﺨﺮﺍﺝ ﺍﻟﺒﻴﺎﻧﺎﺕ ﻏﻴﺮ ﺫﺍﺕ ﺍﻟﺼﻠﺔ ﺃﻭ ﻏﻴﺮ ﺍﻟﻤﺮﻏﻮﺏ‬
‫ﻓﻴﻬﺎ‪ .‬ﻳﺘﻢ ﺫﻟﻚ ﻟﺘﻌﺰﻳﺰ ﻛﻔﺎﺀﺓ ﻭﻓﻌﺎﻟﻴﺔ ﺧﻮﺍﺭﺯﻣﻴﺎﺕ ﺍﻟﺘﻌﻠﻢ ﺍﻵﻟﻲ‪ .‬ﻳﻌﺘﻤﺪ ﺍﻷﺩﺍﺀ ﺍﻟﻔﻌﺎﻝ ﻟﺨﻮﺍﺭﺯﻣﻴﺎﺕ ﺍﻟﺘﻌﻠﻢ ﺍﻵﻟﻲ )‪ (ML‬ﻋﻠﻰ ﺍﻟﻤﻌﺎﻟﺠﺔ‬
‫ﺍﻟﻤﺴﺒﻘﺔ ﺍﻟﻤﻨﺎﺳﺒﺔ ﻟﻤﺠﻤﻮﻋﺔ ﺍﻟﺒﻴﺎﻧﺎﺕ]‪ .[42‬ﺗﺴﺘﺨﺪﻡ ﻫﺬﻩ ﺍﻷﻃﺮﻭﺣﺔ ﺍﻟﻌﺪﻳﺪ ﻣﻦ ﺃﺳﺎﻟﻴﺐ ﺍﻟﻤﻌﺎﻟﺠﺔ ﺍﻟﻤﺴﺒﻘﺔ‪ ،‬ﺑﻤﺎ ﻓﻲ ﺫﻟﻚ ﺗﻮﺍﺯﻥ‬
‫ﺍﻟﺒﻴﺎﻧﺎﺕ‪ ،‬ﻭﺗﻨﻈﻴﻒ ﺍﻟﺒﻴﺎﻧﺎﺕ‪ ،‬ﻭﺗﺮﻣﻴﺰ ﺍﻟﻤﻠﺼﻘﺎﺕ‪ .‬ﺗﻮﻓﺮ ﺍﻷﻗﺴﺎﻡ ﺍﻟﻔﺮﻋﻴﺔ ﺍﻟﻼﺣﻘﺔ ﺷﺮﺣًﺎ ﺷﺎﻣﻼً ﻟﺘﻘﻨﻴﺎﺕ ﺍﻟﻤﻌﺎﻟﺠﺔ ﺍﻟﻤﺴﺒﻘﺔ ﻫﺬﻩ‬
‫‪2.4.1 Data Cleaning‬‬
‫‪The presence of substantial volumes of structured, unstructured, and semi-structured data‬‬
‫‪including noise and anomalies poses challenges in the analysis of such data for the purpose of‬‬
‫‪uncovering novel insights. In addition, the presence of this noise disturbance impedes the‬‬
‫‪cognitive processing of meaning, leading to a suboptimal user experience. Anomalies in data may‬‬
‫‪manifest in several forms, such as the presence of non-standard characters, unstructured‬‬
‫‪punctuation, the use of distinct or multiple languages, misspelt words, and unnecessary columns‬‬
‫‪[43].‬‬

‫‪ 2.4.1‬ﺗﻨﻈﻴﻒ ﺍﻟﺒﻴﺎﻧﺎﺕ‬
‫ﺇﻥ ﻭﺟﻮﺩ ﻛﻤﻴﺎﺕ ﻛﺒﻴﺮﺓ ﻣﻦ ﺍﻟﺒﻴﺎﻧﺎﺕ ﺍﻟﻤﻨﻈﻤﺔ ﻭﻏﻴﺮ ﺍﻟﻤﻨﻈﻤﺔ ﻭﺷﺒﻪ ﺍﻟﻤﻨﻈﻤﺔ ﺑﻤﺎ ﻓﻲ ﺫﻟﻚ ﺍﻟﻀﻮﺿﺎﺀ ﻭﺍﻟﺸﺬﻭﺫﺍﺕ ﻳﺸﻜﻞ ﺗﺤﺪﻳﺎﺕ ﻓﻲ‬
‫ﺗﺤﻠﻴﻞ ﻫﺬﻩ ﺍﻟﺒﻴﺎﻧﺎﺕ ﺑﻐﺮﺽ ﺍﻟﻜﺸﻒ ﻋﻦ ﺭﺅﻯ ﺟﺪﻳﺪﺓ‪ .‬ﺑﺎﻹﺿﺎﻓﺔ ﺇﻟﻰ ﺫﻟﻚ‪ ،‬ﻓﺈﻥ ﻭﺟﻮﺩ ﺍﺿﻄﺮﺍﺏ ﺍﻟﻀﻮﺿﺎﺀ ﻫﺬﺍ ﻳﻌﻴﻖ ﺍﻟﻤﻌﺎﻟﺠﺔ‬
‫ﺍﻟﻤﻌﺮﻓﻴﺔ ﻟﻠﻤﻌﻨﻰ‪ ،‬ﻣﻤﺎ ﻳﺆﺩﻱ ﺇﻟﻰ ﺗﺠﺮﺑﺔ ﻣﺴﺘﺨﺪﻡ ﺩﻭﻥ ﺍﻟﻤﺴﺘﻮﻯ ﺍﻷﻣﺜﻞ‪ .‬ﻗﺪ ﺗﻈﻬﺮ ﺍﻟﺤﺎﻻﺕ ﺍﻟﺸﺎﺫﺓ ﻓﻲ ﺍﻟﺒﻴﺎﻧﺎﺕ ﻓﻲ ﻋﺪﺓ ﺃﺷﻜﺎﻝ‪ ،‬ﻣﺜﻞ‬
‫ﻭﺟﻮﺩ ﺃﺣﺮﻑ ﻏﻴﺮ ﻗﻴﺎﺳﻴﺔ‪ ،‬ﻭﻋﻼﻣﺎﺕ ﺍﻟﺘﺮﻗﻴﻢ ﻏﻴﺮ ﺍﻟﻤﻨﻈﻤﺔ‪ ،‬ﻭﺍﺳﺘﺨﺪﺍﻡ ﻟﻐﺎﺕ ﻣﻤﻴﺰﺓ ﺃﻭ ﻣﺘﻌﺪﺩﺓ‪ ،‬ﻭﻛﻠﻤﺎﺕ ﺑﻬﺎ ﺃﺧﻄﺎﺀ ﺇﻣﻼﺋﻴﺔ‪،‬‬
‫ﻭﺃﻋﻤﺪﺓ ﻏﻴﺮ ﺿﺮﻭﺭﻳﺔ ]‪.[43‬‬

‫‪Figure (2.1) Components of Data Cleaning‬‬

‫‪2.4.2 Label Encoding‬‬

‫‪11‬‬
 Machine learning (ML) is often used in the analysis of datasets that include multiple labels in
one or more columns. The information may consist of numerical or non-numerical data. Certain
classification methods are applicable to categorical variables, whilst others are only suitable for
continuous numerical data. Supervised machine learning algorithms are not designed to handle
word labels, but rather only operate with numeric labels. Hence, the process of assigning
numerical values to label classes is crucial in enabling effective categorization and allowing the
management of datasets via machine learning. The objective of this strategy is to ensure
comprehensibility of all characteristics. This technique may be executed through a range of
methodologies. The two most often used methods for accomplishing this operation are the One
Hot Encoder and the Label Encoder. The Label Encoder is used to encode labels with values
ranging from 0 to m-1, where m is the total number of categories. In the event that a label is
duplicated, it will assign an identical value to that which was assigned on its first occurrence. As
an example, in the case when the dataset has three distinct classes, the label encoding would be
represented by the values 0, 1, and 2. [44].

‫ ﺗﺮﻣﻴﺰ ﺍﻟﻤﻠﺼﻘﺎﺕ‬2.4.2
‫ ﻭﻗﺪ‬.‫( ﻓﻲ ﺗﺤﻠﻴﻞ ﻣﺠﻤﻮﻋﺎﺕ ﺍﻟﺒﻴﺎﻧﺎﺕ ﺍﻟﺘﻲ ﺗﺘﻀﻤﻦ ﺗﺴﻤﻴﺎﺕ ﻣﺘﻌﺪﺩﺓ ﻓﻲ ﻋﻤﻮﺩ ﻭﺍﺣﺪ ﺃﻭ ﺃﻛﺜﺮ‬ML) ‫ﻏﺎﻟﺒًﺎ ﻣﺎ ﻳُﺴﺘﺨﺪﻡ ﺍﻟﺘﻌﻠﻢ ﺍﻵﻟﻲ‬
‫ ﻓﻲ ﺣﻴﻦ ﺃﻥ ﺍﻟﺒﻌﺾ‬،‫ ﺗﻨﻄﺒﻖ ﺑﻌﺾ ﻃﺮﻕ ﺍﻟﺘﺼﻨﻴﻒ ﻋﻠﻰ ﺍﻟﻤﺘﻐﻴﺮﺍﺕ ﺍﻟﻔﺌﻮﻳﺔ‬.‫ﺗﺘﻜﻮﻥ ﺍﻟﻤﻌﻠﻮﻣﺎﺕ ﻣﻦ ﺑﻴﺎﻧﺎﺕ ﺭﻗﻤﻴﺔ ﺃﻭ ﻏﻴﺮ ﺭﻗﻤﻴﺔ‬
‫ ﻟﻢ ﻳﺘﻢ ﺗﺼﻤﻴﻢ ﺧﻮﺍﺭﺯﻣﻴﺎﺕ ﺍﻟﺘﻌﻠﻢ ﺍﻵﻟﻲ ﺍﻟﺨﺎﺿﻌﺔ ﻟﻺﺷﺮﺍﻑ ﻟﻠﺘﻌﺎﻣﻞ ﻣﻊ ﺗﺴﻤﻴﺎﺕ‬.‫ﺍﻵﺧﺮ ﻣﻨﺎﺳﺐ ﻓﻘﻂ ﻟﻠﺒﻴﺎﻧﺎﺕ ﺍﻟﺮﻗﻤﻴﺔ ﺍﻟﻤﺴﺘﻤﺮﺓ‬
‫ ﻓﺈﻥ ﻋﻤﻠﻴﺔ ﺗﻌﻴﻴﻦ ﺍﻟﻘﻴﻢ ﺍﻟﻌﺪﺩﻳﺔ ﻟﻔﺌﺎﺕ ﺍﻟﺘﺼﻨﻴﻒ ﺃﻣﺮ ﺑﺎﻟﻎ ﺍﻷﻫﻤﻴﺔ‬،‫ ﻭﻣﻦ ﺛﻢ‬.‫ ﻭﻟﻜﻨﻬﺎ ﺗﻌﻤﻞ ﻓﻘﻂ ﻣﻊ ﺍﻟﺘﺼﻨﻴﻔﺎﺕ ﺍﻟﺮﻗﻤﻴﺔ‬،‫ﺍﻟﻜﻠﻤﺎﺕ‬
‫ ﺍﻟﻬﺪﻑ ﻣﻦ ﻫﺬﻩ ﺍﻹﺳﺘﺮﺍﺗﻴﺠﻴﺔ ﻫﻮ ﺿﻤﺎﻥ ﻓﻬﻢ‬.‫ﻟﺘﻤﻜﻴﻦ ﺍﻟﺘﺼﻨﻴﻒ ﺍﻟﻔﻌﺎﻝ ﻭﺍﻟﺴﻤﺎﺡ ﺑﺈﺩﺍﺭﺓ ﻣﺠﻤﻮﻋﺎﺕ ﺍﻟﺒﻴﺎﻧﺎﺕ ﻋﺒﺮ ﺍﻟﺘﻌﻠﻢ ﺍﻵﻟﻲ‬
‫ ﺍﻟﻄﺮﻳﻘﺘﺎﻥ ﺍﻷﻛﺜﺮ ﺍﺳﺘﺨﺪﺍﻣًﺎ ﻹﻧﺠﺎﺯ ﻫﺬﻩ ﺍﻟﻌﻤﻠﻴﺔ‬.‫ ﻭﻳﻤﻜﻦ ﺗﻨﻔﻴﺬ ﻫﺬﻩ ﺍﻟﺘﻘﻨﻴﺔ ﻣﻦ ﺧﻼﻝ ﻣﺠﻤﻮﻋﺔ ﻣﻦ ﺍﻟﻤﻨﻬﺠﻴﺎﺕ‬.‫ﺟﻤﻴﻊ ﺍﻟﺨﺼﺎﺋﺺ‬
‫ ﺇﻟﻰ‬0 ‫ ﻳﺘﻢ ﺍﺳﺘﺨﺪﺍﻡ ﺃﺩﺍﺓ ﺗﺸﻔﻴﺮ ﺍﻟﻤﻠﺼﻘﺎﺕ ﻟﺘﺮﻣﻴﺰ ﺍﻟﻤﻠﺼﻘﺎﺕ ﺑﻘﻴﻢ ﺗﺘﺮﺍﻭﺡ ﻣﻦ‬.Label Encoder‫ ﻭ‬One Hot Encoder ‫ﻫﻤﺎ‬
‫ ﺳﻴﺘﻢ ﺗﻌﻴﻴﻦ ﻗﻴﻤﺔ ﻣﻤﺎﺛﻠﺔ ﻟﺘﻠﻚ ﺍﻟﺘﻲ ﺗﻢ ﺗﻌﻴﻴﻨﻬﺎ ﻋﻨﺪ ﻇﻬﻮﺭﻫﺎ‬،‫ ﻓﻲ ﺣﺎﻟﺔ ﺗﻜﺮﺍﺭ ﺍﻟﺘﺴﻤﻴﺔ‬.‫ ﻫﻮ ﺍﻟﻌﺪﺩ ﺍﻹﺟﻤﺎﻟﻲ ﻟﻠﻔﺌﺎﺕ‬m ‫ ﺣﻴﺚ‬،m-1
‫ ﺳﻴﺘﻢ ﺗﻤﺜﻴﻞ ﺗﺮﻣﻴﺰ ﺍﻟﺘﺴﻤﻴﺔ‬،‫ ﻓﻲ ﺍﻟﺤﺎﻟﺔ ﺍﻟﺘﻲ ﺗﺤﺘﻮﻱ ﻓﻴﻬﺎ ﻣﺠﻤﻮﻋﺔ ﺍﻟﺒﻴﺎﻧﺎﺕ ﻋﻠﻰ ﺛﻼﺙ ﻓﺌﺎﺕ ﻣﺘﻤﻴﺰﺓ‬،‫ ﻋﻠﻰ ﺳﺒﻴﻞ ﺍﻟﻤﺜﺎﻝ‬.‫ﻷﻭﻝ ﻣﺮﺓ‬
.[44] .2‫ ﻭ‬1‫ ﻭ‬0 ‫ﺑﺎﻟﻘﻴﻢ‬
2.4.3 Dataset Balancing
One of the primary challenges encountered in machine learning is to the handling of imbalanced
datasets, characterised by a disparity between the classes within the dataset. The introduction of
bias in classification algorithms may significantly diminish the efficacy of classifiers and lead to
12
overfitting issues. Consequently, it becomes imperative to address data imbalance via the process
of data balancing prior to engaging in machine learning [45]. This thesis employs two methods of
dataset balancing, namely over-sampling and under-sampling. The following sections provide a
comprehensive explanation of these two categories:

‫ﻣﻮﺍﺯﻧﺔ ﻣﺠﻤﻮﻋﺔ ﺍﻟﺒﻴﺎﻧﺎﺕ‬.2 .4.3

‫ ﻭﺍﻟﺘﻲ ﺗﺘﻤﻴﺰ ﺑﺎﻟﺘﺒﺎﻳﻦ ﺑﻴﻦ‬،‫ﺃﺣﺪ ﺍﻟﺘﺤﺪﻳﺎﺕ ﺍﻷﺳﺎﺳﻴﺔ ﺍﻟﺘﻲ ﻳﻮﺍﺟﻬﻬﺎ ﺍﻟﺘﻌﻠﻢ ﺍﻵﻟﻲ ﻫﻮ ﺍﻟﺘﻌﺎﻣﻞ ﻣﻊ ﻣﺠﻤﻮﻋﺎﺕ ﺍﻟﺒﻴﺎﻧﺎﺕ ﻏﻴﺮ ﺍﻟﻤﺘﻮﺍﺯﻧﺔ‬
‫ ﻗﺪ ﻳﺆﺩﻱ ﺇﺩﺧﺎﻝ ﺍﻟﺘﺤﻴﺰ ﻓﻲ ﺧﻮﺍﺭﺯﻣﻴﺎﺕ ﺍﻟﺘﺼﻨﻴﻒ ﺇﻟﻰ ﺗﻘﻠﻴﻞ ﻓﻌﺎﻟﻴﺔ ﺍﻟﻤﺼﻨﻔﺎﺕ ﺑﺸﻜﻞ ﻛﺒﻴﺮ ﻭﻳﺆﺩﻱ‬.‫ﺍﻟﻔﺌﺎﺕ ﺩﺍﺧﻞ ﻣﺠﻤﻮﻋﺔ ﺍﻟﺒﻴﺎﻧﺎﺕ‬
‫ ﻳﺼﺒﺢ ﻣﻦ ﺍﻟﻀﺮﻭﺭﻱ ﻣﻌﺎﻟﺠﺔ ﺍﺧﺘﻼﻝ ﺗﻮﺍﺯﻥ ﺍﻟﺒﻴﺎﻧﺎﺕ ﻣﻦ ﺧﻼﻝ ﻋﻤﻠﻴﺔ ﻣﻮﺍﺯﻧﺔ ﺍﻟﺒﻴﺎﻧﺎﺕ‬،‫ ﻭﺑﺎﻟﺘﺎﻟﻲ‬.‫ﺇﻟﻰ ﻣﺸﻜﻼﺕ ﺍﻟﺘﺠﻬﻴﺰ ﺍﻟﺰﺍﺋﺪ‬
‫ ﻭﻫﻤﺎ ﺍﻹﻓﺮﺍﻁ ﻓﻲ ﺃﺧﺬ ﺍﻟﻌﻴﻨﺎﺕ‬،‫ ﺗﺴﺘﺨﺪﻡ ﻫﺬﻩ ﺍﻷﻃﺮﻭﺣﺔ ﻃﺮﻳﻘﺘﻴﻦ ﻟﻤﻮﺍﺯﻧﺔ ﻣﺠﻤﻮﻋﺔ ﺍﻟﺒﻴﺎﻧﺎﺕ‬.[45] ‫ﻗﺒﻞ ﺍﻻﻧﺨﺮﺍﻁ ﻓﻲ ﺍﻟﺘﻌﻠﻢ ﺍﻵﻟﻲ‬
:‫ ﻭﺗﻘﺪﻡ ﺍﻷﻗﺴﺎﻡ ﺍﻟﺘﺎﻟﻴﺔ ﺷﺮﺣﺎ ﺷﺎﻣﻼ ﻟﻬﺎﺗﻴﻦ ﺍﻟﻔﺌﺘﻴﻦ‬.‫ﻭﻧﻘﺺ ﺃﺧﺬ ﺍﻟﻌﻴﻨﺎﺕ‬
A.Under-sampling refers to the procedure of decreasing the representation of the majority class
within a dataset in order to achieve a balanced distribution of classes. This procedure effectively
decreases the quantity of samples inside the majority class, leading to the creation of a dataset that
exhibits a balanced distribution of classes. The aforementioned procedure may be iterated until the
optimal equilibrium is achieved among the various categories within the dataset. Under-sampling
is a beneficial and appropriate technique when the minority classes possess a sufficient number of
samples to adequately carry out the classification process, including both training and testing
phases [46]. This thesis uses the RandomUnderSampling function to execute dataset balancing by
the random elimination of samples from the majority class inside an imbalanced dataset
.‫ ﻳﺸﻴﺮ ﻧﻘﺺ ﺃﺧﺬ ﺍﻟﻌﻴﻨﺎﺕ ﺇﻟﻰ ﺇﺟﺮﺍﺀ ﺗﻘﻠﻴﻞ ﺗﻤﺜﻴﻞ ﻓﺌﺔ ﺍﻷﻏﻠﺒﻴﺔ ﺿﻤﻦ ﻣﺠﻤﻮﻋﺔ ﺑﻴﺎﻧﺎﺕ ﻣﻦ ﺃﺟﻞ ﺗﺤﻘﻴﻖ ﺗﻮﺯﻳﻊ ﻣﺘﻮﺍﺯﻥ ﻟﻠﻔﺌﺎﺕ‬.
‫ ﻣﻤﺎ ﻳﺆﺩﻱ ﺇﻟﻰ ﺇﻧﺸﺎﺀ ﻣﺠﻤﻮﻋﺔ ﺑﻴﺎﻧﺎﺕ ﺗﻌﺮﺽ‬،‫ﻳﺆﺩﻱ ﻫﺬﺍ ﺍﻹﺟﺮﺍﺀ ﺑﺸﻜﻞ ﻓﻌﺎﻝ ﺇﻟﻰ ﺗﻘﻠﻴﻞ ﻛﻤﻴﺔ ﺍﻟﻌﻴﻨﺎﺕ ﺩﺍﺧﻞ ﻓﺌﺔ ﺍﻷﻏﻠﺒﻴﺔ‬
‫ ﻳﻤﻜﻦ ﺗﻜﺮﺍﺭ ﺍﻹﺟﺮﺍﺀ ﺍﻟﻤﺬﻛﻮﺭ ﺃﻋﻼﻩ ﺣﺘﻰ ﻳﺘﻢ ﺗﺤﻘﻴﻖ ﺍﻟﺘﻮﺍﺯﻥ ﺍﻷﻣﺜﻞ ﺑﻴﻦ ﺍﻟﻔﺌﺎﺕ ﺍﻟﻤﺨﺘﻠﻔﺔ ﺩﺍﺧﻞ‬.‫ﺗﻮﺯﻳﻌًﺎ ﻣﺘﻮﺍﺯﻧًﺎ ﻟﻠﻔﺌﺎﺕ‬
‫ ﻳﻌﺪ ﺃﺧﺬ ﺍﻟﻌﻴﻨﺎﺕ ﺍﻟﻨﺎﻗﺺ ﺃﺳﻠﻮﺑًﺎ ﻣﻔﻴﺪًﺍ ﻭﻣﻨﺎﺳﺒًﺎ ﻋﻨﺪﻣﺎ ﺗﻤﺘﻠﻚ ﻓﺌﺎﺕ ﺍﻷﻗﻠﻴﺎﺕ ﻋﺪﺩًﺍ ﻛﺎﻓﻴًﺎ ﻣﻦ ﺍﻟﻌﻴﻨﺎﺕ ﻟﺘﻨﻔﻴﺬ‬.‫ﻣﺠﻤﻮﻋﺔ ﺍﻟﺒﻴﺎﻧﺎﺕ‬
‫ ﺗﺴﺘﺨﺪﻡ ﻫﺬﻩ ﺍﻷﻃﺮﻭﺣﺔ ﻭﻇﻴﻔﺔ‬.[46] ‫ ﺑﻤﺎ ﻓﻲ ﺫﻟﻚ ﻣﺮﺣﻠﺘﻲ ﺍﻟﺘﺪﺭﻳﺐ ﻭﺍﻻﺧﺘﺒﺎﺭ‬،‫ﻋﻤﻠﻴﺔ ﺍﻟﺘﺼﻨﻴﻒ ﺑﺸﻜﻞ ﻣﻨﺎﺳﺐ‬
RandomUnderSampling‫ﻟﺘﻨﻔﻴﺬ ﻣﻮﺍﺯﻧﺔ ﻣﺠﻤﻮﻋﺔ ﺍﻟﺒﻴﺎﻧﺎﺕ ﻋﻦ ﻃﺮﻳﻖ ﺍﻟﺤﺬﻑ ﺍﻟﻌﺸﻮﺍﺋﻲ ﻟﻠﻌﻴﻨﺎﺕ ﻣﻦ ﻓﺌﺔ ﺍﻷﻏﻠﺒﻴﺔ ﺩﺍﺧﻞ‬
‫ﻣﺠﻤﻮﻋﺔ ﺑﻴﺎﻧﺎﺕ ﻏﻴﺮ ﻣﺘﻮﺍﺯﻧﺔ‬

B. Over-sampling refers to the technique used in imbalanced data sets to address the disparity
between minority and majority classes by augmenting the number of samples belonging to the

13
 underrepresented groups until they are equivalent to the majority classes. There are several
strategies that may be used to attain this objective, one of which is Adaptive Synthetic
(ADASYN). The fundamental objective of the ADASYN algorithm is to determine an
appropriate allocation rate for the minority class, which indicates the number of synthetic
samples that need to be created in order to balance the class distribution with the majority
class in the given dataset [47]. ADASYN adheres to a number of mathematical equations,
which are shown below [48]:

‫ ﻳﺸﻴﺮ ﺍﻹﻓﺮﺍﻁ ﻓﻲ ﺃﺧﺬ ﺍﻟﻌﻴﻨﺎﺕ ﺇﻟﻰ ﺍﻟﺘﻘﻨﻴﺔ ﺍﻟﻤﺴﺘﺨﺪﻣﺔ ﻓﻲ ﻣﺠﻤﻮﻋﺎﺕ ﺍﻟﺒﻴﺎﻧﺎﺕ ﻏﻴﺮ ﺍﻟﻤﺘﻮﺍﺯﻧﺔ ﻟﻤﻌﺎﻟﺠﺔ ﺍﻟﺘﻔﺎﻭﺕ ﺑﻴﻦ ﻓﺌﺎﺕ ﺍﻷﻗﻠﻴﺔ‬.‫ﺏ‬
.‫ﻭﺍﻷﻏﻠﺒﻴﺔ ﻣﻦ ﺧﻼﻝ ﺯﻳﺎﺩﺓ ﻋﺪﺩ ﺍﻟﻌﻴﻨﺎﺕ ﺍﻟﺘﻲ ﺗﻨﺘﻤﻲ ﺇﻟﻰ ﺍﻟﻤﺠﻤﻮﻋﺎﺕ ﺍﻟﻤﻤﺜﻠﺔ ﺗﻤﺜﻴﻼ ﻧﺎﻗﺼﺎ ﺣﺘﻰ ﺗﺼﺒﺢ ﻣﻌﺎﺩﻟﺔ ﻟﻄﺒﻘﺎﺕ ﺍﻷﻏﻠﺒﻴﺔ‬
.(ADASYN) ‫ ﻭﺍﺣﺪﺓ ﻣﻨﻬﺎ ﻫﻲ ﺍﻻﺻﻄﻨﺎﻋﻴﺔ ﺍﻟﺘﻜﻴﻔﻴﺔ‬،‫ﻫﻨﺎﻙ ﺍﻟﻌﺪﻳﺪ ﻣﻦ ﺍﻻﺳﺘﺮﺍﺗﻴﺠﻴﺎﺕ ﺍﻟﺘﻲ ﻳﻤﻜﻦ ﺍﺳﺘﺨﺪﺍﻣﻬﺎ ﻟﺘﺤﻘﻴﻖ ﻫﺬﺍ ﺍﻟﻬﺪﻑ‬
‫ ﻭﺍﻟﺬﻱ ﻳﺸﻴﺮ ﺇﻟﻰ ﻋﺪﺩ ﺍﻟﻌﻴﻨﺎﺕ‬،‫ ﻫﻮ ﺗﺤﺪﻳﺪ ﻣﻌﺪﻝ ﺍﻟﺘﺨﺼﻴﺺ ﺍﻟﻤﻨﺎﺳﺐ ﻟﻔﺌﺔ ﺍﻷﻗﻠﻴﺔ‬ADASYN ‫ﺍﻟﻬﺪﻑ ﺍﻷﺳﺎﺳﻲ ﻟﺨﻮﺍﺭﺯﻣﻴﺔ‬
‫ ﺗﻠﺘﺰﻡ‬.[47] ‫ﺍﻻﺻﻄﻨﺎﻋﻴﺔ ﺍﻟﺘﻲ ﻳﺠﺐ ﺇﻧﺸﺎﺅﻫﺎ ﻣﻦ ﺃﺟﻞ ﻣﻮﺍﺯﻧﺔ ﺗﻮﺯﻳﻊ ﺍﻟﻔﺌﺔ ﻣﻊ ﻓﺌﺔ ﺍﻷﻏﻠﺒﻴﺔ ﻓﻲ ﻣﺠﻤﻮﻋﺔ ﺍﻟﺒﻴﺎﻧﺎﺕ ﺍﻟﻤﺤﺪﺩﺓ‬
:[48] ‫ ﺑﻌﺪﺩ ﻣﻦ ﺍﻟﻤﻌﺎﺩﻻﺕ ﺍﻟﺮﻳﺎﺿﻴﺔ ﺍﻟﻤﺒﻴﻨﺔ ﺃﺩﻧﺎﻩ‬ADASYN

1) Using Equation (2.1), can be measure the Degree ( ᵅᵅᵅ ) at which the minority classes are
related to the majority classes:
ᵃᵅᵅᵅ
ᵅᵅᵅ = ……………………………………………(2.1)
ᵃᵅᵄᵆ
Where ᵃᵅᵄᵆ represents samples of the larger group of classes and ᵃᵅᵄᵅ represents
samples of the minority classes.
2) The computation of synthetic samples may be achieved by using Equation (2.2).
ᵃ = (ᵃᵅᵄᵆ − ᵃᵅᵅᵅ ) * ᵯ …….………………..…………..(2.2)
The variable ᵯ represents the degree of balance in synthetic samples, whereas ᵃ represents
the amount of synthetic samples.

3) The KNN algorithm is used to compute the dominance of the majority classes, as shown by
Equation (2.3).
14
 ᵃᵅᵅ = ∆ ᵅᵅ / ᵃ ………………………………………………(2.3)
The variable ᵃᵅ is the measure of dominance of the majority classes inside a particular
neighbourhood, whereas ∆ᵅ i represents the number of samples in the KNN algorithm. The
variable ᵅ i denotes the quantity of minority class samples, whereas ᵃ represents the count
of nearest neighbours.
4) Equation (2.4) is used for calculating the amount of synthetic samples for every
neighbourhood:
ᵃᵅᵅ = ᵃ * ᵃᵅᵅ ……………………………………………(2.4)
where the amount of synthetic samples for every neighbourhood is represented by ᵃᵅ i.
1) Equation (2.5) is used to produce novel synthetic samples for every minority class:
ᵄᵅᵅᵆ = ᵄᵅᵅ + (ᵄᵆᵅᵅ − ᵄᵅᵅ ) * ᵰ ……………………………..(2.5)
Where Xᵆᵅ i and Xᵅ i denote two minority samples residing inside the same
neighbourhood, and is a random variable that takes values between 0 and 1.
The act of selecting a random subset from several categories is often known as random under-
sampling. This methodology is iteratively used until the state of equilibrium is attained within the
minority group. While this approach facilitates the achievement of equilibrium between the
majority and minority groups, the exclusion of samples from the majority group.

Figure (2.2) Dataset after under sampling [49]

The method of obtaining samples from a minority community in a random and non-
discriminatory manner is sometimes referred to as random oversampling. The iterative process is

15
continued until the imbalanced majority class is rectified; nevertheless, it is important to note that
this approach has the risk of inducing overfitting of the training model to the minority class.

Figure (2.3) Dataset after oversampling [49]

2.2.4 Feature Selection

Some ML algorithms require higher computational time in the classification process if the dataset
volume is large; therefore; only the necessary features should be selected from the dataset to build
the model, ignoring the unimportant data. The feature selection process eliminates redundant and
unimportant values while ensuring that valuable information is not lost because such irrelevant
data leads to high computational cost and need more time. Many feature selection techniques,
including SelectKBest, use a chi-square scoring function. The scoring function identifies the best
features by examining the degree of correlation between the features and labels in the dataset and
selecting the features with the highest chi-square score as the best influential features in the
classification process. The value of the chi-square is calculated in Equation (2.6) [50]. The
determination of the number of target features is contingent upon the value assigned to the pre-
established parameter k.

(
s2 ∑ (( ᵅᵅ E
O –
ᵅᵅ
Eᵅᵅ
)
)) ………………………………….(2.6)

16
Where the ᵂ ᵽ is chi-square value, ᵇ ᵈᵈ is the observed value, and ᵀ ᵈᵈ is the expected value and
the ᵈ ᵈ number of classes.

2.5 Artificial Intelligence

Artificial Intelligence (AI) is a prominent field within the discipline of computer science,
exerting significant influence across several domains of study. AI has the capability to acquire
information and engage in logical thinking processes in order to address intricate challenges. AI
encompasses the examination and advancement of intelligent devices and software systems that
possess the ability to acquire knowledge, process information, manipulate data, engage in
communication, and comprehend various things. AI technologies provide several advantages
across a wide range of applications. The primary domains of artificial intelligence are[51] [52]:

 Expert Systems
 Speech Understanding
 Robotics and Sensory Systems
 Natural Language Processing
 Computer Vision and Scene Recognition
 Neural Computing
 Intelligent Computer-Aided Instruction
Artificial intelligence (AI) is increasingly being used across several domains to address intricate
challenges and facilitate decision-making processes. The use of this measure guarantees both cost-
effectiveness and dependability. While AI offers significant advantages in terms of efficient
problem-solving, it also presents some limitations by potentially displacing human workers in
several domains, hence resulting in unemployment[53]. Artificial intelligence primarily relies on
approaches that are developed based on scientific principles, including statistics, mathematics, and
biology. Several methodologies often used in the field of artificial intelligence include Neural
Networks (NN), Evolutionary Computing (EC), Genetic Algorithms (GA), Fuzzy Logic (FL), and
Hybrid Artificial Intelligence (HAI). [54][55].

2.6 Machine Learning

17
Machine learning is a specialised domain within the science of artificial intelligence that focuses
on the formulation and implementation of methodologies and approaches that enable
computational systems to acquire knowledge and improve performance via experience. Machine
learning is a field of study that focuses on the development of algorithms that enable machines to
acquire knowledge and perform jobs and activities. This pertains to the development of a
computer programme that enhances its performance via experiential learning. The machine
learning system undergoes training using a sample set of data in order to enhance its performance
when used to execute the required function, depending on the acquired learning. In recent times,
several pragmatic applications of machine learning have been procured, including data-mining
software, autonomous automobiles, pattern recognition systems, and information filtering systems
[56].
The following are some areas where machine learning algorithms excel and stand out [57]:
 Problems with lengthy lists of possible adjustments or needing a great deal of human
intervention to solve.
 Rules: One machine learning approach may be used to increase speed and simplify the
code.
 Complex problems that cannot be solved optimally using typical methods.
 Methodology: The use of machine learning methods enables the identification of the most
effective solution.
 Changing Conditions: A machine learning system is able to adjust to different
environments.
 Gaining comprehensive understanding of intricate matters and huge amounts of data.
2.6.1 Machine Learning Types
ML is a computational approach within the field of artificial intelligence that involves the analysis
of data and the construction of a model. ML approaches enable the system to do data analysis and
acquire knowledge via learning. Furthermore, after the training process on the dataset, the system
has the capability to discern patterns, forecast future outcomes, and execute accurate decision-
making. ML systems may be classified according on the level and kind of supervision they get
throughout the training process. There are four primary types, as seen in Figure (2.4) [57] [58] :
18
 Figure (2.4) illustrates the basic types of machine learning [58].

1. Supervised Learning: The computer undergoes supervised training, wherein it is provided

with both input data and corresponding output data. This training process enables the
computer to establish connections and identify patterns between the input data and output
data, thereby equipping it with the ability to make predictions for new input data in the
future. Supervised learning often involves doing tasks like as classification and regression,
where the former involves categorising data into distinct classes, while the latter entails
predicting a numerical value referred to as the target variable.
2. Unsupervised Learning: The computer undergoes training solely through the provision of
data, hence referred to as unsupervised learning, as there is no supervision during the
learning process and no output is provided to guide the data. Through this training, the
computer establishes connections and identifies patterns within the data, enabling it to make
predictions about future data outputs. Common problems in unsupervised learning include
grouping and dimensionality reduction. Figure (2.5) shows how supervised learning is
different from unsupervised learning.

19
 Figure (2 .5): The difference between unsupervised learning and supervised learning

3. Semi-supervised learning: One approach that combines the capabilities of supervised and
unsupervised learning methods. This approach integrates both the labelled and unlabeled
data to construct an appropriate classifier or function.
4. Reinforcement Learning: is a model for machine learning. The learning system, referred
to as the agent within this particular context, undergoes training via a process based on
rewards and punishments. The agent engages in the process of observing its surroundings,
making deliberate selections of actions, and then executing those activities in order to get
rewards for appropriate acts or face penalties for inappropriate ones. The agent strives to
maximise the number of accurate actions while minimising the occurrence of erroneous
actions. Figure (2.6) show reinforcement learning.

20
 Figure (2.6): Reinforcement learning [57]

2.6.2 Identification Algorithms in Machine Learning

A selection of the many machine learning classification and identification techniques will be
succinctly discussed in this section.

A. Decision Tree Algorithm (DT)

The use of a widely employed decision tree strategy is a proficient approach for creating
models in the format of a hierarchical tree structure [59]. The Decision Tree (DT) approach is
used to partition a dataset into progressively smaller subgroups. Simultaneously, the construction
of a decision tree is being carried out incrementally. This approach produces a hierarchical
structure consisting of nodes and leaves to represent decision-making processes. Decision trees
have the capability to handle both numerical and categorical input in their processing. Figure (2.7)
illustrates the core concept of the decision tree. The decision tree rule is represented by Equation
2.7[60].

ᵅ −1 . . . . . . . . . . . . . . . . . . . . .. . . . . . (2.7)
ᵃᵅᵆᵅᵅᵅᵆ (ᵆ ) = − ∑ᵅ (ᵅ |ᵆ ) ᵅᵅᵅ
ᵅ =0
2 ᵅ (ᵅ |ᵆ )

Where c represent the total number of classes. Additionally, (ᵅ │ᵆ ) is used to represent the
proportion of records that belong to class i at a certain node t.

21
 Figure (2.7 ): illustrates the core concept of the DT.

The following terms apply to the decision trees shown in picture 2.7: [60]

 Root Node: It divides further into two or more homogenous groups and reflects the full
population or sample.
 Splitting: During this procedure, a node is divided into two or more sub-nodes.
 Decision Node: whenever an existing sub node splits up into new ones
 Leaf/ Terminal Node: Nodes that aren't divided (Leaf or Terminal nodes)
 Pruning: when the sub-nodes of a decision node are removed.
 Branch (Sub-Tree): A branch or sub-tree is a smaller branch that makes up the overall
tree.
 Parent and Child Node: The term "parent node of sub-nodes" refers to a node that has
sub-nodes.
B. Random Forest
The Random Forest (RF) technique is a widely used approach for both regression and
classification tasks. It involves training several decision tree classifiers on different subsets of a
given dataset. The construction of RF is based on the decision tree as described in reference [61].
A multitude of trees are constructed and afterwards validated via the use of a training set, with the

22
purpose of making predictions for future observations. The output of this approach has the
potential to be either categorical or continuous in nature.

The fundamental principle behind this technique is the creation of a limited number of features
via the construction of tiny decision trees, which are computationally efficient procedures. If it
were possible to construct several tiny decision trees concurrently, the resulting trees might be
merged into a single robust learner by a combination or majority voting approach [62]. Figure
(2.8) illustrates the core concept of the RF method.

Figure (2.8): illustrates the core concept of the RF method.

The Random Forest approach has many advantages. Firstly, it is utilizedto both classification
and regression tasks. Secondly, it demonstrates robustness in handling missing values, hence
preserving accuracy even when confronted with incomplete data. Lastly, it exhibits efficacy in
managing huge datasets with high dimensionality. [63][64].

During the training phase, the classification process of a Random Forest (RF) algorithm
involves each tree learning from a randomly selected subset of the data points. The outcomes of
classification trees are subsequently condensed. In the testing phase, when a new input sample is
introduced, each decision tree inside the forest independently classifies the sample into a specific
category using a classification algorithm. Additionally, a voting mechanism is used to choose the

23
most specific category for the sample, enabling the forest to make accurate predictions for that
particular category.

C. Logistic Regression (LR)

Logistic regression (LR) is a statistical method used to estimate the likelihood of an event
taking place, based on the values of variables that are associated or may be linked to that
occurrence. Logistic regression employs a set of predictor variables, which may be either
numerical or categorical in nature. This modelling technique is extensively used in several
scientific and commercial domains, making it one of the most prevalent strategies employed in the
area of machine learning. This observation illustrates a correlation between a set Y of groups and
a feature-vector X. This is accomplished by using a logistic regression approach to ascertain the
association between the class and the feature vector. The assumption is made that the distribution
P(Y|X), where Y represents the class and X represents the feature-vector, has a boundary-like
form. Subsequently, this assumption is validated by analysing the training data. The probability
P(Y|X) of X belonging to class Y is determined by the sigmoidal function, as seen in equations
(2.8) and (2.9) [65]
ᵇ

ᵃ (ᵁ , ᵁ ) = ∑ᵉ
ᵈ =1
ᵈ ᵈ ᵈ (ᵈ , ᵇ ) ……………………………………………… (2.8)

1 ………………………………………………..… (2.9)
ᵉ ( ᵈ |ᵇ ) =
1 + ᵁᵃᵂ⁡ ( − ᵉ (ᵈ , ᵇ ) )

where:
P is the probability of (Y|X).
Y is representing the class.
X is representing the feature-vector.
w is the weight of the word.
exp is the exponential.
f is the frequency.

24
 P (Y|X) is shown by multiplying the attributes X by a certain weight wi, then linearly
connecting them, and applying a function fi (Y, X) on the correlations. fi is a function that
establishes a connection between a class characteristic and the characteristic in a feature-vector,
classifying it as true or false depending on whether the probability above a certain threshold. The
weight of a quality reflects its "strength" since certain traits are more important than others.[65].
Multinomial logistic regression, a version of this technique, may also be used to categorize
more than two targets. The function produces a result of 0 or 1, which reflects the negative (0) and
the positive (1) situation in the binary method as show in Figure 2.9.

Figure (2.9): logistic regression method

D. Support Vector Machine (SVM)

The Support Vector Machine (SVM) is a common Supervised Learning technique for both
classification and regression tasks. Nevertheless, its primary use is Classification tasks. The SVM
algorithm's objective is to find the optimal line or decision boundary that divides the n-
dimensional space into classes, making it simple to assign the new data point to the right class in
the future. This optimal decision boundary is referred to as a hyperplane.

25
 SVM selects the extremal points/vectors that contribute to the formation of the hyperplane.
The name "Support Vector Machine" refers to the fact that this technique is designed to help in
the most severe instances.
In binary classification, the hyperplane separates dataset points into two classes. As for multi-
classes classification, it is the division of a multi-classes classification problem into several
binary-class classification problems, and hyperplane is used to separate a multi-classes dataset by
using the approach of One-Vs- All (OVA) or One-Vs-One (OVO) [66]. Classification using the
OVO strategy is to divide the multi-class classification into multiple binary classifications, and the
number of binary classifiers is determined by Equation (2.10) [67]:

ᵄ =
ᵅᵅ (ᵅᵅ − 1) ................................................................(2.10)
2

Where ᵇ is several binary classification problems, and the ᵈᵈ is the number of classes.
Each binary classifier predicts a specific class for the new test sample, which selects the
appropriate class by voting, meaning that the classes that get the most votes are the classes for the
new test sample [66]. Classification is done in an OVO approach based on the voting approach as
follows: if the first pair of classes is (a,b) and the new test sample to be classified as on the (b)
side of the hyperplane, b will take one vote, as for the second pair of classes (a,c) if the test
sample is on the c side of the hyperplane, then class c gets one vote, if the third pair is (b,c) and
the test sample is on the b side of the hyperplane, then class b gets two votes, and class (c) gets
one vote, the third class (a) gets zero votes. The test sample is classified as (b). as shown in Figure
(2.10) [68].

26
 Figure (2.10) Multi-class Classification with SVM [68]
The hyperplane is calculated in Equation (2.11) [69] :

ᵆ .ᵆ +ᵅ =0 ………………………………. (2.11)

Where ᵉ is a data point called a support vector to be located if it is to the left or right of the
hyperplane, and (ᵉ , ᵉ ) are parameters of the hyperplane. In addition, the SVM algorithm uses
various mathematical functions called the kernel that uses data as input and then converts that data
to the required algorithm. There are different kernel types: the linear, Radial Basis Function
(RBF), polynomial, and sigmoid kernel [70]. The stages depicted in figure 2.10 that make up the
random forest algorithm are:

 Step 1: Random samples are selected by the procedure from the given dataset.
 Step 2: A decision tree will be produced for each sample selected by the algorithm.
The anticipated outcome will then be provided by each decision tree.
 Step 3: Following that, voting will be held for each anticipated result. It will use mean
for a regression problem and mode for a classification difficulty.
 Step 4: Finally, the system will choose the final prediction with the highest votes.
E. K-Nearest Neighbor (KNN)

This algorithm is one of the supervised ML algorithms and is considered more

straightforward than the others. Since learning in this algorithm is supervised, the samples in each
vector must be named for training samples. Then, in the classification phase determining the test
sample for the classes to which the sample belongs. The distance between the training sample
points and the new sample point must be calculated, and the ᵅ value that represents the depth of
the algorithm must be determined, that is, the number of points in the classification process used
in the classification process [71]. Finally, the appropriate classes are selected for each new sample
point, depending on most of the neighbor’s votes [71]. Depending on the number of classes in the
dataset, the dataset is split into a certain number of neighbourhoods, as illustrates in Figure (2.11).

27
 Figure (2.11) illustrates example of the K-Nearest neighbours algorithm[72]
The value ᵈ must be determined before starting the classification because it determines the
scope of training (number of neighbours) according to the number of samples present. The
distance among the new data point and the training data points for each neighbour is measured
using different measurements, including the 28uclidean and 28uclidean distance measurements.
The Euclidean distance is measured by Equation (2.12) [73].

ᵆ ………………………………….(2.12)
ᵅ (ᵆ , ᵆ ) = ∑(ᵆ + ᵆ
ᵅ =1
ᵅ ) 2

Whereᵈ is the distance among the new data point and the training data points, theᵉ is the
number of the training sample, the ᵉ new data points, and the ᵉ is the training data points; the
new data point is assigned to the class with the most significant number of nearest neighbours.
The following method may be used to describe how the K-NN works:

 Step-1: Elect the number of the Kth neighbour.

 Step-2: compute the Euclidean distance among K neighbours.
 Step-3: Using the Euclidean distance calculation, choose the K nearest neighbours.
 Step-4: Determine the number of data points in each of these k neighbours.
 Step-5: Select the category with the largest neighbour count, and then allocate the extra data
points to it.
28
  Step-6: Stop.
2.7 Performance Measures
The system uses many metrics with the purpose of assessing and quantifying system
performance. The evaluation metrics include accuracy, F1-score, precision, and recall. The
computation of these metrics relies on the use of a confusion matrix, which is a matrix that
provides a summary of the correct and incorrect predictions made by a classification model, as
shown in Table (2.1).

The primary values of this Table are delineated herein[74]

Table (2.1): Confusion matrix

Actual Values

Positive Negative

Positive TP FP
Predicated
Values
Negative FN TN

 True Positive (TP): This term refers to the positive cases that show a correct categorization.
 False Negative (FN): This term refers to the positive cases that show a falsely
categorization.
 False Positive (FP): This term refers to the negative cases that falsely categorization.
 True Negative (TN): This term refers to the negative cases that a correct categorization.
This section described the metrics of accuracy, precision, recall and F1-score:

A. Precision: is a metric that quantifies the proportion of true positives correctly identified,
relative to the sum of true positives and false positives. The accuracy may be calculated using
equation (2.13) [74].

ᵇᵇ …………………………………………………... (2.13)
ᵀᵂᵁᵁᵂᵂᵂᵂᵂ =
ᵇᵇ + ᵆᵇ
29
B. Accuracy: refers to the amount of accurate forecasts, calculated by dividing the number of
correct predictions by the total number of predictions made. The accuracy of the computation may
be measured by using Equation (2.14)[74].

ᵇᵇ + ᵇᵇ ………………………………………….. (2.14)
ᵆᵈᵈᵉᵉᵈᵈᵉ =
ᵇᵇ + ᵇᵇ + ᵆᵇ + ᵆᵇ

C. Recall: Also referred to as sensitivity, recall quantifies the number of positive examples that
the model correctly classifies; a high recall value indicates that there are relatively few positive
cases that are mistakenly labelled as negatives, as illustrates in Equation (2.15) [74].

ᵇᵇ ………………………………………………..…… (2.15)
ᵇᵈᵈᵈᵈᵈ =
ᵇᵇ + ᵆᵇ

D. F-measure: The F-measure, often known as the F-score, is a metric that combines precision
and recall to evaluate the performance of a test or classification system. precision is defined as the
ratio of the number of true positive results to the total number of positive outcomes. The recall
metric is defined as the ratio of true positive results to the total number of actual positive that were
expected to be retrieved. The conventional or balanced F1 score (mean F1) is the average
harmonic value of precision and recall. F1 score reaches its ideal value of one and deteriorates
towards zero, as seen in Equation (2.16) [74].

1 ᵅᵅᵅᵅᵅᵆᵅᵅᵅ * ᵅᵅᵅᵄᵅᵅ ………………………….. (2.16)

ᵆ1= 1 1
=2*
+ ᵅᵅᵅᵅᵅᵆᵅᵅᵅ + ᵅᵅᵅᵄᵅᵅ
ᵅᵅᵅᵄᵅᵅ ᵅᵅᵅᵅᵅᵆᵅᵅᵅ

E. Error Rate: It calculates the ratio of incorrectly predicted samples to the total predictions of
samples in the dataset. The best value for the error rate is zero and is calculated in Equation (2.17)
[74].

ᵃᵄ + ᵃᵄ ………………………………. (2.17)
ᵃᵅᵅᵅᵅ ᵄᵄᵆᵅ = * 100 %
ᵄᵄ + ᵃᵄ + ᵃᵄ + ᵄᵄ

F. ROC: The receiver operating characteristic (ROC) curve is frequently used for evaluating the
performance of binary classification algorithms. It provides a graphical representation of a
classifier’s performance, rather than a single value like most other metrics. The ROC curve is

30
produced by calculating and plotting the True Positive Rate (TPR) against the False Positive
Rate (FPR) for a single

31