Abstract
This action research study examined the use of an AI chatbot as a learning support tool for engine fault diagnosis among 30 first-semester Malaysian Vocational Certificate (SVM) Automotive Technology students at a vocational college in Selangor. It followed the Kemmis and McTaggart action research model using a one-group pre-test and post-test design across a seven-week study period. The chatbot was not a sole diagnostic source: every response had to be verified against the service manual, physical inspection and teacher confirmation under a layered verification protocol. The mean score rose from 39.10 (SD = 3.42) to 82.70 (SD = 5.91), a difference of 43.60 points, 95% CI [41.37, 45.83], t(29) = 39.93, p < .001, with a normalized gain of 0.716. All 30 participants improved, but only 19 reached the high N-Gain category. The Cohen's dz of 7.29 is not interpreted as an intervention effect: the narrow pre-test standard deviation and the absence of a control group mean it reflects sample homogeneity and ordinary curriculum learning. A 20-item questionnaire recorded an overall mean of 4.31 (SD = 0.32) with a Cronbach's alpha of .897, 95% CI [.835, .943]; however, 96% of the 600 responses fell on scale points 4 or 5 and none below 3, a pattern consistent with socially desirable responding given that the researcher was also the class teacher. Findings are reported as changes in achievement and acceptance following the intervention, not as causal evidence. The study's contribution is a layered verification protocol for TVET diagnostic teaching.
Abstrak: Kajian tindakan ini meneliti penggunaan AI Chatbot sebagai alat sokongan pembelajaran diagnostik kerosakan enjin dalam kalangan 30 orang pelajar Sijil Vokasional Malaysia (SVM) Teknologi Automotif Semester 1 di sebuah kolej vokasional di Selangor. Kajian dilaksanakan mengikut model kajian tindakan Kemmis dan McTaggart menggunakan reka bentuk satu kumpulan ujian pra dan ujian pos dalam tempoh kajian tujuh minggu. AI Chatbot tidak digunakan sebagai sumber diagnosis tunggal kerana setiap respons wajib disahkan melalui manual servis, pemeriksaan fizikal dan bimbingan guru mengikut protokol pengesahan berlapis. Skor min meningkat daripada 39.10 (SP = 3.42) kepada 82.70 (SP = 5.91), iaitu perbezaan 43.60 mata, 95% SK [41.37, 45.83], t(29) = 39.93, p < .001, dengan peningkatan ternormal 0.716. Kesemua 30 peserta merekodkan peningkatan, tetapi hanya 19 orang mencapai kategori N-Gain tinggi. Saiz kesan Cohen dz sebanyak 7.29 tidak ditafsirkan sebagai kesan intervensi kerana sisihan piawai ujian pra yang sempit dan ketiadaan kumpulan kawalan menjadikan nilai tersebut mencerminkan kehomogenan sampel dan pembelajaran kurikulum biasa. Soal selidik 20 item merekodkan min keseluruhan 4.31 (SP = 0.32) dengan alfa Cronbach .897, 95% SK [.835, .943], namun 96% daripada 600 respons berada pada skala 4 atau 5 dan tiada satu pun respons di bawah 3, iaitu pola yang konsisten dengan kecenderungan jawapan yang diingini secara sosial memandangkan pengkaji juga guru kelas. Dapatan dilaporkan sebagai perubahan pencapaian dan penerimaan selepas intervensi, bukan bukti kausal. Sumbangan utama kajian ialah protokol pengesahan berlapis bagi pengajaran diagnostik TVET.